Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,8 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+output/finetuned_model/checkpoint-100/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+output/finetuned_model/checkpoint-165/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+output/finetuned_model/checkpoint-33/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+output/finetuned_model/final_model/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+output/finetuned_model/merged_model/tokenizer.json filter=lfs diff=lfs merge=lfs -text

Makefile

@@ -0,0 +1,44 @@
+# Makefile
+.PHONY: all config analyze generate train merge evaluate clean
+
+all: config analyze generate train merge evaluate
+
+config:
+	@echo "Step 0: Updating configuration..."
+	python utils/config_manager.py
+
+analyze:
+	@echo "Step 1: Analyzing repository..."
+	python scripts/01_analyze_repo.py
+
+generate:
+	@echo "Step 2: Generating training data..."
+	python scripts/02_generate_data.py
+
+train:
+	@echo "Step 3: Fine-tuning model..."
+	deepspeed --num_gpus=2 scripts/03_train_model.py
+
+merge:
+	@echo "Step 4: Merging LoRA weights..."
+	python scripts/04_merge_weights.py
+
+evaluate:
+	@echo "Step 5: Evaluating model..."
+	python scripts/05_evaluate.py
+
+clean:
+	@echo "Cleaning output files..."
+	rm -rf output/finetuned_model/checkpoints/*
+	rm -rf data/training_data/*
+
+help:
+	@echo "Available targets:"
+	@echo "  make all       - Run complete pipeline"
+	@echo "  make config    - Update repository config"
+	@echo "  make analyze   - Analyze code repository"
+	@echo "  make generate  - Generate training data"
+	@echo "  make train     - Fine-tune model"
+	@echo "  make merge     - Merge LoRA weights"
+	@echo "  make evaluate  - Evaluate model"
+	@echo "  make clean     - Clean output files"

README.md

@@ -1143,7 +1143,3 @@ python scripts/05_evaluate.py --quick-eval
 - 企业内部代码助手
 - 开源项目文档生成
 - 代码审查辅助
-
----
-license: mit
----

config/deepspeed_zero3.json

@@ -0,0 +1,38 @@
+{
+  "bf16": {
+    "enabled": true
+  },
+  "zero_optimization": {
+    "stage": 3,
+    "offload_optimizer": {
+      "device": "cpu",
+      "pin_memory": true
+    },
+    "offload_param": {
+      "device": "cpu",
+      "pin_memory": true
+    },
+    "overlap_comm": true,
+    "contiguous_gradients": true,
+    "sub_group_size": 1e9,
+    "reduce_bucket_size": "auto",
+    "stage3_prefetch_bucket_size": "auto",
+    "stage3_param_persistence_threshold": "auto",
+    "stage3_max_live_parameters": 1e9,
+    "stage3_max_reuse_distance": 1e9,
+    "stage3_gather_16bit_weights_on_model_save": true
+  },
+  "gradient_accumulation_steps": "auto",
+  "gradient_clipping": "auto",
+  "steps_per_print": 10,
+  "train_batch_size": "auto",
+  "train_micro_batch_size_per_gpu": "auto",
+  "wall_clock_breakdown": false,
+  "aio": {
+    "block_size": 1048576,
+    "queue_depth": 8,
+    "thread_count": 1,
+    "single_submit": false,
+    "overlap_events": true
+  }
+}

config/default_config.yaml

@@ -0,0 +1,87 @@
+data_generation:
+  design_proposals:
+    count: 100
+    requirement_types:
+    - 新功能开发
+    - 性能优化
+    - 架构重构
+    - API设计
+    - 错误处理
+  qa_pairs:
+    count: 500
+    diversity_threshold: 0.7
+    max_code_lines: 40
+    min_code_lines: 5
+dataset:
+  format: jsonl
+  output_dir: ./data/training_data
+  test_split: 0.1
+  train_split: 0.8
+  val_split: 0.1
+evaluation:
+  metrics:
+  - rouge
+  - bleu
+  - exact_match
+  sample_size: 50
+gpu:
+  devices:
+  - 0
+  - 1
+  memory_per_gpu: 48
+llm_api:
+  batch_size: 4
+  max_workers: 2
+  model: Qwen/Qwen3-8B
+  provider: local
+model:
+  base_model: Qwen/Qwen3-8B
+  enable_thinking: true
+  max_length: 2048
+  temperature: 0.7
+  thinking_budget: 4096
+  top_p: 0.9
+project:
+  name: code_repo_training_data_generator
+  version: 1.0.0
+repository:
+  exclude_dirs:
+  - .git
+  - __pycache__
+  - node_modules
+  - .venv
+  - venv
+  - build
+  - dist
+  languages:
+  - python
+  - markdown
+  local_path: ./repos/Laddr
+  url: https://github.com/AgnetLabs/Laddr
+training:
+  batch_size: 2
+  bf16: true
+  deepspeed_config: ./deepspeed_config_optimized.json
+  eval_steps: 100
+  gradient_accumulation_steps: 8
+  learning_rate: 1e-3
+  logging_steps: 10
+  lora:
+    alpha: 128
+    bias: none
+    dropout: 0.05
+    r: 64
+    target_modules:
+    - q_proj
+    - k_proj
+    - v_proj
+    - o_proj
+    - gate_proj
+    - up_proj
+    - down_proj
+  max_grad_norm: 1.0
+  num_epochs: 3
+  output_dir: ./output/finetuned_model
+  save_steps: 100
+  warmup_ratio: 0.05
+  weight_decay: 0.01

data/training_data/metadata.json

@@ -0,0 +1,15 @@
+{
+  "total_samples": 440,
+  "train_samples": 352,
+  "val_samples": 44,
+  "test_samples": 44,
+  "project_name": "Laddr",
+  "task_distribution": {
+    "code_explanation": 212,
+    "code_location": 75,
+    "api_usage": 150,
+    "project_structure": 1,
+    "project_overview": 1,
+    "core_components": 1
+  }
+}

data/training_data/test.jsonl

@@ -0,0 +1,44 @@
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`_parse_ts` 函数在哪个文件里？"}, {"role": "assistant", "content": "`_parse_ts` 位于 `lib/laddr/src/laddr/api/main.py`。"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`ps` 有什么功能？"}, {"role": "assistant", "content": "`ps` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/commands/management.py`。\n**功能描述**：\nShow status of all services and workers. Examples: laddr ps laddr ps --format json\n**函数签名**：\n```python\ndef ps(format: str):\n```\n**参数**：\n- `format` (str)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`run_prompt` 函数在哪个文件里？"}, {"role": "assistant", "content": "`run_prompt` 位于 `lib/laddr/src/laddr/cli/commands/prompt.py`。"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `MCPToolSource` 是做什么的？"}, {"role": "assistant", "content": "`MCPToolSource` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/mcp_client.py`。\n**功能描述**：\nPlaceholder - MCP feature disabled for this release.\n**函数签名**：\n```python\nclass MCPToolSource:\n```"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `DatabaseService.memory_put` 是做什么的？"}, {"role": "assistant", "content": "`DatabaseService.memory_put` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/database.py`。\n**功能描述**：\nStore a memory entry.\n**函数签名**：\n```python\ndef memory_put(self, agent_name: str, key: str, value: Any, job_id: str | None = None) -> None:\n```\n**参数**：\n- `self` (Any)\n- `agent_name` (str)\n- `key` (str)\n- `value` (Any)\n- `job_id` (str | None)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`build_image` 有什么功能？"}, {"role": "assistant", "content": "`build_image` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/utils/docker.py`。\n**功能描述**：\nBuild a Docker image. Args: dockerfile: Path to Dockerfile tag: Image tag context: Build context directory Raises: DockerComposeError: If build fails\n**函数签名**：\n```python\ndef build_image(dockerfile: Path, tag: str, context: Path) -> None:\n```\n**参数**：\n- `dockerfile` (Path): Path to Dockerfile\n- `tag` (str)\n- `context` (Path)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `BackendFactory` 的作用。"}, {"role": "assistant", "content": "`BackendFactory` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/config.py`。\n**功能描述**：\nFactory for creating backend instances based on configuration. Provides pluggable implementations for queue, database, LLM, and cache.\n**函数签名**：\n```python\nclass BackendFactory:\n```"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`compose_logs` 有什么功能？"}, {"role": "assistant", "content": "`compose_logs` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/utils/docker.py`。\n**功能描述**：\nView logs for a Docker Compose service. Args: service: Service name follow: Follow log output (default: False) tail: Number of lines to show from end (default: all) Raises: DockerComposeError: If compose logs fails\n**函数签名**：\n```python\ndef compose_logs(\nservice: str,\nfollow: bool = False,\ntail: int | None = None,\n) -> None:\n```\n**参数**：\n- `service` (str): Service name\n- `follow` (bool)\n- `tail` (int | None)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`ProjectDetails.validate_name` 有什么功能？"}, {"role": "assistant", "content": "`ProjectDetails.validate_name` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/cli/utils/config.py`。\n**功能描述**：\nValidate project name format.\n**函数签名**：\n```python\ndef validate_name(cls, v: str) -> str:\n```\n**参数**：\n- `cls` (Any)\n- `v` (str)\n\n**返回值**：`str`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `BackendFactory.create_database_backend` 函数？"}, {"role": "assistant", "content": "`BackendFactory.create_database_backend` 位于 `lib/laddr/src/laddr/core/config.py`，使用方法如下：\n**参数说明**：\n\n**功能简述**：Create database backend based on config...."}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`tool` 有什么功能？"}, {"role": "assistant", "content": "`tool` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/core/tooling.py`。\n**功能描述**：\nDecorator to mark a function as an agent tool. Usage: @tool(name=\"web_search\", description=\"Search the web\") def search(query: str) -> dict: return {\"results\": [...]} Or with Pydantic input model: class SearchInput(BaseModel): query: str limit: int = 10 @tool() def search(inputs: SearchInput) -> dict: return {\"results\": [...]} The decorator attaches metadata to the function as __laddr_tool__.\n**函数签名**：\n```python\ndef tool(\nname: str | None = None,\ndescription: str | None = None,\n*,\ntrace: bool = True,\n```\n**参数**：\n- `name` (str | None)\n- `description` (str | None)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `ToolRegistry.__init__` 函数？"}, {"role": "assistant", "content": "`ToolRegistry.__init__` 位于 `lib/laddr/src/laddr/core/tooling.py`，使用方法如下：\n**参数说明**："}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `decorator` 函数？"}, {"role": "assistant", "content": "`decorator` 位于 `lib/laddr/src/laddr/core/system_tools.py`，使用方法如下：\n```python\ndecorator(func=...)\n```\n**参数说明**：\n- `func`: Callable"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `run` 的作用。"}, {"role": "assistant", "content": "`run` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/commands/run.py`。\n**功能描述**：\nRun environments, agents, or pipelines. Usage: laddr run dev [--build] [--no-detach] laddr run agent <agent_name> [--inputs '{...}'] laddr run pipeline <file.yml>\n**函数签名**：\n```python\ndef run(ctx: click.Context):\n```\n**参数**：\n- `ctx` (click.Context)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`AgentRunner.replay` 方法在哪个文件里？"}, {"role": "assistant", "content": "`AgentRunner.replay` 位于 `lib/laddr/src/laddr/core/runtime_entry.py`。"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `nearest_whitespace` 函数？"}, {"role": "assistant", "content": "`nearest_whitespace` 位于 `lib/laddr/src/laddr/core/system_tools.py`，使用方法如下：\n```python\nnearest_whitespace(target=..., low=..., high=...)\n```\n**参数说明**：\n- `target`: int\n- `low`: int\n- `high`: int"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`prompt` 函数在哪个文件里？"}, {"role": "assistant", "content": "`prompt` 位于 `lib/laddr/src/laddr/cli/commands/prompt.py`。"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `OllamaLLM.__init__` 函数？"}, {"role": "assistant", "content": "`OllamaLLM.__init__` 位于 `lib/laddr/src/laddr/core/llm.py`，使用方法如下：\n```python\nOllamaLLM.__init__(base_url=..., model=...)\n```\n**参数说明**：\n- `base_url`: str | None\n- `model`: str | None"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`GrokLLM` 类在哪个文件里？"}, {"role": "assistant", "content": "`GrokLLM` 位于 `lib/laddr/src/laddr/core/llm.py`。"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`MCPToolRegistry` 类在哪个文件里？"}, {"role": "assistant", "content": "`MCPToolRegistry` 位于 `lib/laddr/src/laddr/core/mcp_client.py`。"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `RedisCache.__init__` 函数？"}, {"role": "assistant", "content": "`RedisCache.__init__` 位于 `lib/laddr/src/laddr/core/cache.py`，使用方法如下：\n```python\nRedisCache.__init__(redis_url=...)\n```\n**参数说明**：\n- `redis_url`: str"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `InMemoryCache.delete` 函数？"}, {"role": "assistant", "content": "`InMemoryCache.delete` 位于 `lib/laddr/src/laddr/core/cache.py`，使用方法如下：\n```python\nInMemoryCache.delete(key=...)\n```\n**参数说明**：\n- `key`: str\n\n**功能简述**：Delete value from cache...."}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`ProjectConfig` 有什么功能？"}, {"role": "assistant", "content": "`ProjectConfig` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/config.py`。\n**功能描述**：\nProject configuration from laddr.yml. Defines which agents exist and backend preferences.\n**函数签名**：\n```python\nclass ProjectConfig(BaseModel):\n```"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `EventThrottler` 的作用。"}, {"role": "assistant", "content": "`EventThrottler` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/api/main.py`。\n**功能描述**：\nThrottle WebSocket events to avoid overwhelming clients.\n**函数签名**：\n```python\nclass EventThrottler:\n```"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `AgentRunner._discover_local_agents` 函数？"}, {"role": "assistant", "content": "`AgentRunner._discover_local_agents` 位于 `lib/laddr/src/laddr/core/runtime_entry.py`，使用方法如下：\n```python\nAgentRunner._discover_local_agents(exclude=...)\n```\n**参数说明**：\n- `exclude`: list[str] | None\n\n**功能简述**：Discover local agents under the 'agents' package for inline execution...."}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`AgentRunner.__init__` 有什么功能？"}, {"role": "assistant", "content": "`AgentRunner.__init__` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/runtime_entry.py`。\n**功能描述**：\nInitialize runner. Args: agent: Agent instance (created if None) env_config: Environment configuration\n**函数签名**：\n```python\ndef __init__(\nself,\nagent: Agent | None = None,\nenv_config: LaddrConfig | None = None\n):\n```\n**参数**：\n- `self` (Any)\n- `agent` (Agent | None): Agent instance (created if None)\n- `env_config` (LaddrConfig | None)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`add_tool` 函数在哪个文件里？"}, {"role": "assistant", "content": "`add_tool` 位于 `lib/laddr/src/laddr/cli/commands/add.py`。"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `ps` 函数？"}, {"role": "assistant", "content": "`ps` 位于 `lib/laddr/src/laddr/cli/commands/management.py`，使用方法如下：\n```python\nps(format=...)\n```\n**参数说明**：\n- `format`: str\n\n**功能简述**：Show status of all services and workers. Examples: laddr ps laddr ps --format json..."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `discover_tools` 函数？"}, {"role": "assistant", "content": "`discover_tools` 位于 `lib/laddr/src/laddr/core/tooling.py`，使用方法如下：\n```python\ndiscover_tools(agent_name=...)\n```\n**参数说明**：\n- `agent_name`: str - Name of the agent (e.g., \"researcher\")\n\n**功能简述**：Auto-discover tools for an agent from agents.<agent_name>.tools package. Scans for: 1. Functions decorated with @tool 2. Functions named 'run' (fallback) Args: agent_name: Name of the agent (e.g., \"re..."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `MCPToolRegistry.__init__` 函数？"}, {"role": "assistant", "content": "`MCPToolRegistry.__init__` 位于 `lib/laddr/src/laddr/core/mcp_client.py`，使用方法如下：\n**参数说明**："}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`Agent` 有什么功能？"}, {"role": "assistant", "content": "`Agent` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/agent_runtime.py`。\n**功能描述**：\nBase Agent class with full runtime capabilities. Provides: - Planning and execution - Tool calling with validation and caching - Delegation with blocking wait for responses - Memory storage - Auto-registration on message bus - Trace logging to database Subclasses should override plan() method for custom planning logic, or use the default autonomous_run() for intelligent LLM-driven execution.\n**函数签名**：\n```python\nclass Agent:\n```"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `run` 函数？"}, {"role": "assistant", "content": "`run` 位于 `lib/laddr/src/laddr/cli/commands/run.py`，使用方法如下：\n```python\nrun(ctx=...)\n```\n**参数说明**：\n- `ctx`: click.Context\n\n**功能简述**：Run environments, agents, or pipelines. Usage: laddr run dev [--build] [--no-detach] laddr run agent <agent_name> [--inputs '{...}'] laddr run pipeline <file.yml>..."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `print_header` 函数？"}, {"role": "assistant", "content": "`print_header` 位于 `lib/laddr/src/laddr/cli/utils/logger.py`，使用方法如下：\n```python\nprint_header(text=..., style=...)\n```\n**参数说明**：\n- `text`: str - Header text\n- `style`: str\n\n**功能简述**：Print a minimalistic header. Args: text: Header text style: Text style (default: bold cyan)..."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `ArtifactStorageTool.__init__` 函数？"}, {"role": "assistant", "content": "`ArtifactStorageTool.__init__` 位于 `lib/laddr/src/laddr/core/system_tools.py`，使用方法如下：\n```python\nArtifactStorageTool.__init__(storage_backend=..., default_bucket=...)\n```\n**参数说明**：\n- `storage_backend`: Any - Storage backend (MinIO/S3)\n- `default_bucket`: str | None\n\n**功能简述**：Initialize artifact storage tool. Args: storage_backend: Storage backend (MinIO/S3)..."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `ollama` 函数？"}, {"role": "assistant", "content": "`ollama` 位于 `lib/laddr/src/laddr/llms.py`，使用方法如下：\n```python\nollama(model=..., temperature=..., base_url=...)\n```\n**参数说明**：\n- `model`: str | None\n- `temperature`: float | None\n- `base_url`: str | None\n\n**功能简述**：Create a local Ollama backend instance. Example: llm = ollama(model=\"gemma2:2b\", base_url=\"http://localhost:11434\")..."}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`anthropic` 函数在哪个文件里？"}, {"role": "assistant", "content": "`anthropic` 位于 `lib/laddr/src/laddr/llms.py`。"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `compose_scale` 是做什么的？"}, {"role": "assistant", "content": "`compose_scale` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/utils/docker.py`。\n**功能描述**：\nScale a Docker Compose service. Args: service: Service name replicas: Number of replicas Raises: DockerComposeError: If scaling fails\n**函数签名**：\n```python\ndef compose_scale(service: str, replicas: int) -> None:\n```\n**参数**：\n- `service` (str): Service name\n- `replicas` (int)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `ToolRegistry.register` 的作用。"}, {"role": "assistant", "content": "`ToolRegistry.register` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/tooling.py`。\n**功能描述**：\nRegister a tool with optional aliases. Args: tool_obj: Tool instance or callable to register name: Optional explicit name (overrides tool_obj.name) aliases: Optional list of alias names that resolve to this tool\n**函数签名**：\n```python\ndef register(self, tool_obj: Tool | Callable, name: str | None = None, aliases: list[str] | None = None) -> None:\n```\n**参数**：\n- `self` (Any)\n- `tool_obj` (Tool | Callable): Tool instance or callable to register\n- `name` (str | None)\n- `aliases` (list[str] | None)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `GroqLLM.__init__` 函数？"}, {"role": "assistant", "content": "`GroqLLM.__init__` 位于 `lib/laddr/src/laddr/core/llm.py`，使用方法如下：\n```python\nGroqLLM.__init__(api_key=..., model=...)\n```\n**参数说明**：\n- `api_key`: str | None\n- `model`: str | None"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `AnthropicLLM.__init__` 函数？"}, {"role": "assistant", "content": "`AnthropicLLM.__init__` 位于 `lib/laddr/src/laddr/core/llm.py`，使用方法如下：\n```python\nAnthropicLLM.__init__(api_key=..., model=...)\n```\n**参数说明**：\n- `api_key`: str | None\n- `model`: str | None"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`AnthropicLLM` 有什么功能？"}, {"role": "assistant", "content": "`AnthropicLLM` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/llm.py`。\n**功能描述**：\nAnthropic LLM backend.\n**函数签名**：\n```python\nclass AnthropicLLM:\n```"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`print_header` 有什么功能？"}, {"role": "assistant", "content": "`print_header` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/utils/logger.py`。\n**功能描述**：\nPrint a minimalistic header. Args: text: Header text style: Text style (default: bold cyan)\n**函数签名**：\n```python\ndef print_header(text: str, style: str = \"bold cyan\") -> None:\n```\n**参数**：\n- `text` (str): Header text\n- `style` (str)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `print_completion` 函数？"}, {"role": "assistant", "content": "`print_completion` 位于 `lib/laddr/src/laddr/cli/utils/logger.py`，使用方法如下：\n```python\nprint_completion(message=...)\n```\n**参数说明**：\n- `message`: str - Completion message\n\n**功能简述**：Print a completion message with checkmark. Args: message: Completion message..."}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `S3Storage._get_client` 是做什么的？"}, {"role": "assistant", "content": "`S3Storage._get_client` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/storage.py`。\n**功能描述**：\nGet synchronous S3-compatible client.\n**函数签名**：\n```python\ndef _get_client(self):\n```\n**参数**：\n- `self` (Any)\n\n**返回值**：`None`"}]}

data/training_data/val.jsonl

@@ -0,0 +1,44 @@
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `print_step` 的作用。"}, {"role": "assistant", "content": "`print_step` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/utils/logger.py`。\n**功能描述**：\nPrint a step in a process with minimalistic style. Args: step: Step name/action description: Optional description\n**函数签名**：\n```python\ndef print_step(step: str, description: str = \"\") -> None:\n```\n**参数**：\n- `step` (str): Step name/action\n- `description` (str)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`print_completion` 有什么功能？"}, {"role": "assistant", "content": "`print_completion` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/utils/logger.py`。\n**功能描述**：\nPrint a completion message with checkmark. Args: message: Completion message\n**函数签名**：\n```python\ndef print_completion(message: str) -> None:\n```\n**参数**：\n- `message` (str): Completion message\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`Agent._format_history_for_synthesis` 有什么功能？"}, {"role": "assistant", "content": "`Agent._format_history_for_synthesis` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/agent_runtime.py`。\n**功能描述**：\nFormat execution history for final answer synthesis.\n**函数签名**：\n```python\ndef _format_history_for_synthesis(self, history: list[dict]) -> str:\n```\n**参数**：\n- `self` (Any)\n- `history` (list[dict])\n\n**返回值**：`str`"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `AgentMemory.list` 是做什么的？"}, {"role": "assistant", "content": "`AgentMemory.list` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/agent_runtime.py`。\n**功能描述**：\nList all memory entries.\n**函数签名**：\n```python\ndef list(self) -> dict[str, Any]:\n```\n**参数**：\n- `self` (Any)\n\n**返回值**：`dict[str, Any]`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `gemini` 函数？"}, {"role": "assistant", "content": "`gemini` 位于 `lib/laddr/src/laddr/llms.py`，使用方法如下：\n```python\ngemini(model=..., temperature=...)\n```\n**参数说明**：\n- `model`: str | None\n- `temperature`: float | None"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`TemplateRenderer.__init__` 方法在哪个文件里？"}, {"role": "assistant", "content": "`TemplateRenderer.__init__` 位于 `lib/laddr/src/laddr/cli/utils/templates.py`。"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`DatabaseService.get_trace` 有什么功能？"}, {"role": "assistant", "content": "`DatabaseService.get_trace` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/database.py`。\n**功能描述**：\nGet a single trace event by id with full payload.\n**函数签名**：\n```python\ndef get_trace(self, trace_id: str) -> dict | None:\n```\n**参数**：\n- `self` (Any)\n- `trace_id` (str)\n\n**返回值**：`dict | None`"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `Memory` 是做什么的？"}, {"role": "assistant", "content": "`Memory` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/database.py`。\n**功能描述**：\nAgent memory storage.\n**函数签名**：\n```python\nclass Memory(Base):\n```"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`ps` 函数在哪个文件里？"}, {"role": "assistant", "content": "`ps` 位于 `lib/laddr/src/laddr/cli/commands/management.py`。"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `ServiceNotReadyError` 是做什么的？"}, {"role": "assistant", "content": "`ServiceNotReadyError` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/cli/utils/errors.py`。\n**功能描述**：\nRaised when a service is not ready after startup.\n**函数签名**：\n```python\nclass ServiceNotReadyError(LaddrError):\n```"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `list_prompts` 是做什么的？"}, {"role": "assistant", "content": "`list_prompts` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/commands/prompt.py`。\n**功能描述**：\nList recent prompt executions.\n**函数签名**：\n```python\ndef list_prompts(limit: int):\n```\n**参数**：\n- `limit` (int)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `DatabaseBackend.save_result` 函数？"}, {"role": "assistant", "content": "`DatabaseBackend.save_result` 位于 `lib/laddr/src/laddr/core/config.py`，使用方法如下：\n```python\nDatabaseBackend.save_result(job_id=..., outputs=...)\n```\n**参数说明**：\n- `job_id`: str\n- `outputs`: dict"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `OllamaLLM` 的作用。"}, {"role": "assistant", "content": "`OllamaLLM` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/llm.py`。\n**功能描述**：\nLocal Ollama LLM backend (talks to a local Ollama HTTP server). Default base URL: http://localhost:11434 Expected model names: e.g. \"gemma2:2b\" (use LLM_MODEL or per-agent LLM_MODEL_<AGENT>)\n**函数签名**：\n```python\nclass OllamaLLM:\n```"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `MemoryBus.__init__` 函数？"}, {"role": "assistant", "content": "`MemoryBus.__init__` 位于 `lib/laddr/src/laddr/core/message_bus.py`，使用方法如下：\n**参数说明**："}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `add_tool` 函数？"}, {"role": "assistant", "content": "`add_tool` 位于 `lib/laddr/src/laddr/cli/commands/add.py`，使用方法如下：\n```python\nadd_tool(tool_name=..., agent=..., description=...)\n```\n**参数说明**：\n- `tool_name`: str\n- `agent`: str | None\n- `description`: str | None\n\n**功能简述**：Add a new tool to an agent. Examples: laddr add tool calculator laddr add tool scraper --agent researcher..."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `BackendFactory.create_cache_backend` 函数？"}, {"role": "assistant", "content": "`BackendFactory.create_cache_backend` 位于 `lib/laddr/src/laddr/core/config.py`，使用方法如下：\n**参数说明**：\n\n**功能简述**：Create cache backend based on config...."}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`AgentRegistry` 有什么功能？"}, {"role": "assistant", "content": "`AgentRegistry` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/database.py`。\n**功能描述**：\nAgent registration and metadata.\n**函数签名**：\n```python\nclass AgentRegistry(Base):\n```"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `ProjectDetails.validate_name` 函数？"}, {"role": "assistant", "content": "`ProjectDetails.validate_name` 位于 `lib/laddr/src/laddr/cli/utils/config.py`，使用方法如下：\n```python\nProjectDetails.validate_name(cls=..., v=...)\n```\n**参数说明**：\n- `cls`: Any\n- `v`: str\n\n**功能简述**：Validate project name format...."}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`TemplateRenderer` 类在哪个文件里？"}, {"role": "assistant", "content": "`TemplateRenderer` 位于 `lib/laddr/src/laddr/cli/utils/templates.py`。"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `MemoryBus` 是做什么的？"}, {"role": "assistant", "content": "`MemoryBus` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/message_bus.py`。\n**功能描述**：\nIn-memory message bus for testing and quick local dev. Not suitable for production or distributed systems.\n**函数签名**：\n```python\nclass MemoryBus:\n```"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `Agent.__init__` 的作用。"}, {"role": "assistant", "content": "`Agent.__init__` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/agent_runtime.py`。\n**功能描述**：\nInitialize agent. Args: config: Agent-specific configuration env_config: Environment configuration (loaded if None) tools: Tool registry (auto-discovered if None)\n**函数签名**：\n```python\ndef __init__(\nself,\nconfig: AgentConfig,\nenv_config: LaddrConfig | None = None,\ntools: ToolRegistry | list[Any] | None = None,\n```\n**参数**：\n- `self` (Any)\n- `config` (AgentConfig): Agent-specific configuration\n- `env_config` (LaddrConfig | None)\n- `tools` (ToolRegistry | list[Any] | None)\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `Tool.validate_inputs` 函数？"}, {"role": "assistant", "content": "`Tool.validate_inputs` 位于 `lib/laddr/src/laddr/core/tooling.py`，使用方法如下：\n```python\nTool.validate_inputs(inputs=...)\n```\n**参数说明**：\n- `inputs`: dict[str, Any]\n\n**功能简述**：Validate inputs against input_model if available. Returns validated inputs or raises ValidationError...."}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `bind_tools` 是做什么的？"}, {"role": "assistant", "content": "`bind_tools` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/core/tooling.py`。\n**功能描述**：\nBind tools to an agent instance for explicit, readable tool registration. Supports: - String names: auto-import from agents.<agent>.tools.<name> - Callables: register directly - Decorated functions: extract Tool metadata Usage: from agents.researcher.tools import web_search bind_tools(self, [web_search.run, \"summarize\"]) Args: agent_instance: Agent instance with .tools registry tools: List of tool names (str) or callables\n**函数签名**：\n```python\ndef bind_tools(agent_instance: Any, tools: list[str | Callable]) -> None:\n```\n**参数**：\n- `agent_instance` (Any): Agent instance with .tools registry\n- `tools` (list[str | Callable])\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`gemini` 函数在哪个文件里？"}, {"role": "assistant", "content": "`gemini` 位于 `lib/laddr/src/laddr/llms.py`。"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `init` 函数？"}, {"role": "assistant", "content": "`init` 位于 `lib/laddr/src/laddr/cli/commands/init.py`，使用方法如下：\n```python\ninit(project_name=..., path=...)\n```\n**参数说明**：\n- `project_name`: str | None\n- `path`: str\n\n**功能简述**：Initialize a new Laddr project. Creates project structure with: - Configuration files (laddr.yml, .env) - Docker setup (docker-compose.yml, Dockerfile) - Default researcher agent - Example tools and p..."}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`AgentChatRequest` 有什么功能？"}, {"role": "assistant", "content": "`AgentChatRequest` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/api/main.py`。\n**功能描述**：\nRequest to chat with an agent.\n**函数签名**：\n```python\nclass AgentChatRequest(BaseModel):\n```"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`AgentMemory.get` 有什么功能？"}, {"role": "assistant", "content": "`AgentMemory.get` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/agent_runtime.py`。\n**功能描述**：\nRetrieve a memory entry.\n**函数签名**：\n```python\ndef get(self, key: str) -> Any:\n```\n**参数**：\n- `self` (Any)\n- `key` (str)\n\n**返回值**：`Any`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `DatabaseService.list_prompts` 函数？"}, {"role": "assistant", "content": "`DatabaseService.list_prompts` 位于 `lib/laddr/src/laddr/core/database.py`，使用方法如下：\n```python\nDatabaseService.list_prompts(limit=...)\n```\n**参数说明**：\n- `limit`: int\n\n**功能简述**：List recent prompt executions...."}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`wrap_with_runtime_injection` 函数在哪个文件里？"}, {"role": "assistant", "content": "`wrap_with_runtime_injection` 位于 `lib/laddr/src/laddr/core/system_tools.py`。"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `_print_service_info` 的作用。"}, {"role": "assistant", "content": "`_print_service_info` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/commands/run.py`。\n**功能描述**：\nPrint service URLs in minimalistic style.\n**函数签名**：\n```python\ndef _print_service_info() -> None:\n```\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`ProjectDetails.validate_name` 方法在哪个文件里？"}, {"role": "assistant", "content": "`ProjectDetails.validate_name` 位于 `lib/laddr/src/laddr/cli/utils/config.py`。"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`PipelineStage` 有什么功能？"}, {"role": "assistant", "content": "`PipelineStage` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/config.py`。\n**功能描述**：\nStage definition in a pipeline (simplified).\n**函数签名**：\n```python\nclass PipelineStage(BaseModel):\n```"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `_build_trace_tree` 的作用。"}, {"role": "assistant", "content": "`_build_trace_tree` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/api/main.py`。\n**功能描述**：\nBuild hierarchical trace tree from flat traces. Groups by agent runs and nests tool calls/LLM calls within them.\n**函数签名**：\n```python\ndef _build_trace_tree(traces: list[dict]) -> list[dict]:\n```\n**参数**：\n- `traces` (list[dict])\n\n**返回值**：`list[dict]`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 项目中，`Agent._parse_autonomous_action` 有什么功能？"}, {"role": "assistant", "content": "`Agent._parse_autonomous_action` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/agent_runtime.py`。\n**功能描述**：\nParse LLM response to extract action, preferring JSON format with text fallback.\n**函数签名**：\n```python\ndef _parse_autonomous_action(self, response: str) -> dict:\n```\n**参数**：\n- `self` (Any)\n- `response` (str)\n\n**返回值**：`dict`"}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `print_warning` 的作用。"}, {"role": "assistant", "content": "`print_warning` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/utils/logger.py`。\n**功能描述**：\nPrint a warning message. Args: message: Warning message\n**函数签名**：\n```python\ndef print_warning(message: str) -> None:\n```\n**参数**：\n- `message` (str): Warning message\n\n**返回值**：`None`"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `DatabaseService.get_job_traces` 函数？"}, {"role": "assistant", "content": "`DatabaseService.get_job_traces` 位于 `lib/laddr/src/laddr/core/database.py`，使用方法如下：\n```python\nDatabaseService.get_job_traces(job_id=...)\n```\n**参数说明**：\n- `job_id`: str\n\n**功能简述**：Get all traces for a job...."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `DatabaseService.register_agent` 函数？"}, {"role": "assistant", "content": "`DatabaseService.register_agent` 位于 `lib/laddr/src/laddr/core/database.py`，使用方法如下：\n```python\nDatabaseService.register_agent(agent_name=..., metadata=...)\n```\n**参数说明**：\n- `agent_name`: str\n- `metadata`: dict\n\n**功能简述**：Register or update an agent in the registry...."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `DatabaseService.get_result` 函数？"}, {"role": "assistant", "content": "`DatabaseService.get_result` 位于 `lib/laddr/src/laddr/core/database.py`，使用方法如下：\n```python\nDatabaseService.get_result(job_id=...)\n```\n**参数说明**：\n- `job_id`: str\n\n**功能简述**：Get job result...."}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `CacheBackend` 是做什么的？"}, {"role": "assistant", "content": "`CacheBackend` 是 Laddr 项目中的一个 类，位于 `lib/laddr/src/laddr/core/config.py`。\n**功能描述**：\nProtocol for cache backends.\n**函数签名**：\n```python\nclass CacheBackend(Protocol):\n```"}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `Agent._trace` 函数？"}, {"role": "assistant", "content": "`Agent._trace` 位于 `lib/laddr/src/laddr/core/agent_runtime.py`，使用方法如下：\n```python\nAgent._trace(job_id=..., agent_name=..., event_type=..., payload=...)\n```\n**参数说明**：\n- `job_id`: str\n- `agent_name`: str\n- `event_type`: str\n- `payload`: dict\n\n**功能简述**：Append a trace if enabled and not masked...."}]}
+{"conversations": [{"role": "user", "content": "如何在 Laddr 中使用 `Agent._parse_autonomous_action_text` 函数？"}, {"role": "assistant", "content": "`Agent._parse_autonomous_action_text` 位于 `lib/laddr/src/laddr/core/agent_runtime.py`，使用方法如下：\n```python\nAgent._parse_autonomous_action_text(response=...)\n```\n**参数说明**：\n- `response`: str\n\n**功能简述**：Parse legacy text format with ACTION:/AGENT:/TOOL_NAME:/... lines...."}]}
+{"conversations": [{"role": "user", "content": "请解释 Laddr 中 `DatabaseService.create_prompt` 的作用。"}, {"role": "assistant", "content": "`DatabaseService.create_prompt` 是 Laddr 项目中的一个 方法，位于 `lib/laddr/src/laddr/core/database.py`。\n**功能描述**：\nCreate a new prompt execution record.\n**函数签名**：\n```python\ndef create_prompt(self, prompt_id: str | None, prompt_name: str, inputs: dict) -> str:\n```\n**参数**：\n- `self` (Any)\n- `prompt_id` (str | None)\n- `prompt_name` (str)\n- `inputs` (dict)\n\n**返回值**：`str`"}]}
+{"conversations": [{"role": "user", "content": "在 Laddr 中，`DatabaseBackend.list_agents` 方法在哪个文件里？"}, {"role": "assistant", "content": "`DatabaseBackend.list_agents` 位于 `lib/laddr/src/laddr/core/config.py`。"}]}
+{"conversations": [{"role": "user", "content": "Laddr 的 `print_table` 是做什么的？"}, {"role": "assistant", "content": "`print_table` 是 Laddr 项目中的一个 函数，位于 `lib/laddr/src/laddr/cli/utils/logger.py`。\n**功能描述**：\nPrint data as a rich table. Args: data: List of dictionaries (each dict is a row) title: Optional table title\n**函数签名**：\n```python\ndef print_table(data: list[dict], title: str | None = None) -> None:\n```\n**参数**：\n- `data` (list[dict]): List of dictionaries (each dict is a row)\n- `title` (str | None)\n\n**返回值**：`None`"}]}

output/finetuned_model/all_results.json

@@ -0,0 +1,12 @@
+{
+    "epoch": 15.0,
+    "eval_loss": 0.36164844036102295,
+    "eval_runtime": 19.5939,
+    "eval_samples_per_second": 2.246,
+    "eval_steps_per_second": 0.561,
+    "total_flos": 4194690293760.0,
+    "train_loss": 0.13649824394731586,
+    "train_runtime": 4647.838,
+    "train_samples_per_second": 1.136,
+    "train_steps_per_second": 0.036
+}

output/finetuned_model/checkpoint-100/README.md

@@ -0,0 +1,207 @@
+---
+base_model: Qwen/Qwen3-8B
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3-8B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.17.1

output/finetuned_model/checkpoint-100/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3-8B",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "down_proj",
+    "o_proj",
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

output/finetuned_model/checkpoint-100/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4bdb15b8b2fddfc07564cb9cc2628910883a425bca84faaa179995f91121f28
+size 349244256

output/finetuned_model/checkpoint-100/added_tokens.json

@@ -0,0 +1,28 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

output/finetuned_model/checkpoint-100/chat_template.jinja

@@ -0,0 +1,89 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = '' %}
+    {%- endif %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
+            {%- endif %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

output/finetuned_model/checkpoint-100/global_step100/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5db0035a11255731a939f91310dd0b203605027b8612e596757b9c0c9302768b
+size 1047531685

output/finetuned_model/checkpoint-100/global_step100/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40f28549882323943a7e3fb546378acb36aef4a8afa5be819e9a463439522eb2
+size 1047531685

output/finetuned_model/checkpoint-100/global_step100/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad8435c4684810b01226e6bd9356601fd67e07d616aea9fc706100f90778f351
+size 504721

output/finetuned_model/checkpoint-100/global_step100/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8ea3e44dab4fe4bd0dc8c9d8b7e1b69fa0664df796447980d7d69185bb8182e
+size 504721

output/finetuned_model/checkpoint-100/latest

@@ -0,0 +1 @@
+global_step100

output/finetuned_model/checkpoint-100/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11feaaeae1609ce17295b2ede1dd07125882006761ebeb4d125c8ba0a00acf6b
+size 14917

output/finetuned_model/checkpoint-100/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a16d6396806fbc784f6e7bb03f2a7b4a6f569c964ccf5fe1bfadf6389e05af8f
+size 14917

output/finetuned_model/checkpoint-100/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bedc55c2b89b8c70fee4c419e7d755566357dcab6a912093210f03702324e13
+size 1465

output/finetuned_model/checkpoint-100/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

output/finetuned_model/checkpoint-100/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
+size 11422654

output/finetuned_model/checkpoint-100/tokenizer_config.json

@@ -0,0 +1,240 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

output/finetuned_model/checkpoint-100/trainer_state.json

@@ -0,0 +1,119 @@
+{
+  "best_global_step": 100,
+  "best_metric": 0.36164844036102295,
+  "best_model_checkpoint": "output/finetuned_model/checkpoint-100",
+  "epoch": 9.090909090909092,
+  "eval_steps": 100,
+  "global_step": 100,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.09090909090909091,
+      "grad_norm": 4.3420233726501465,
+      "learning_rate": 0.0,
+      "loss": 2.8404,
+      "step": 1
+    },
+    {
+      "epoch": 0.9090909090909091,
+      "grad_norm": 0.5920354127883911,
+      "learning_rate": 0.0005,
+      "loss": 1.3885,
+      "step": 10
+    },
+    {
+      "epoch": 1.8181818181818183,
+      "grad_norm": 0.37487199902534485,
+      "learning_rate": 0.0004949476630105669,
+      "loss": 0.3859,
+      "step": 20
+    },
+    {
+      "epoch": 2.7272727272727275,
+      "grad_norm": 0.4673846364021301,
+      "learning_rate": 0.0004799948609147061,
+      "loss": 0.1584,
+      "step": 30
+    },
+    {
+      "epoch": 3.6363636363636362,
+      "grad_norm": 0.2643900513648987,
+      "learning_rate": 0.0004557459664734141,
+      "loss": 0.0623,
+      "step": 40
+    },
+    {
+      "epoch": 4.545454545454545,
+      "grad_norm": 0.21961501240730286,
+      "learning_rate": 0.00042318108837739987,
+      "loss": 0.0477,
+      "step": 50
+    },
+    {
+      "epoch": 5.454545454545454,
+      "grad_norm": 0.14172837138175964,
+      "learning_rate": 0.00038361645653195025,
+      "loss": 0.0259,
+      "step": 60
+    },
+    {
+      "epoch": 6.363636363636363,
+      "grad_norm": 0.05771714448928833,
+      "learning_rate": 0.0003386512217606339,
+      "loss": 0.0128,
+      "step": 70
+    },
+    {
+      "epoch": 7.2727272727272725,
+      "grad_norm": 0.04812893271446228,
+      "learning_rate": 0.0002901028202144401,
+      "loss": 0.0076,
+      "step": 80
+    },
+    {
+      "epoch": 8.181818181818182,
+      "grad_norm": 0.03544061630964279,
+      "learning_rate": 0.0002399335149726463,
+      "loss": 0.0046,
+      "step": 90
+    },
+    {
+      "epoch": 9.090909090909092,
+      "grad_norm": 0.010216760449111462,
+      "learning_rate": 0.00019017108392811063,
+      "loss": 0.0036,
+      "step": 100
+    },
+    {
+      "epoch": 9.090909090909092,
+      "eval_loss": 0.36164844036102295,
+      "eval_runtime": 19.4544,
+      "eval_samples_per_second": 2.262,
+      "eval_steps_per_second": 0.565,
+      "step": 100
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 165,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 15,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2537229926400.0,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

output/finetuned_model/checkpoint-100/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26f3e70da891d4b4c03100e0214975b3f0494b364a6dd21c49136d030c4a1005
+size 7825

output/finetuned_model/checkpoint-100/zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+#   python zero_to_fp32.py . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device, weights_only=False)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+    total_files = len(files)
+    state_dicts = []
+    for f in tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                                          tag,
+                                                          exclude_frozen_parameters,
+                                                          lazy_mode=True)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info("Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info("Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument("output_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)

output/finetuned_model/checkpoint-165/README.md

@@ -0,0 +1,207 @@
+---
+base_model: Qwen/Qwen3-8B
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3-8B
+- lora
+- transformers
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.17.1

output/finetuned_model/checkpoint-165/adapter_config.json

@@ -0,0 +1,42 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3-8B",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "down_proj",
+    "o_proj",
+    "gate_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

output/finetuned_model/checkpoint-165/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3967b5fe0e7ae6170e9f8d2af8dad8d9c1488c73d5724ef1db8c1ddbba1de2f0
+size 349244256

output/finetuned_model/checkpoint-165/added_tokens.json

@@ -0,0 +1,28 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

output/finetuned_model/checkpoint-165/chat_template.jinja

@@ -0,0 +1,89 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = '' %}
+    {%- endif %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
+            {%- endif %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

output/finetuned_model/checkpoint-165/global_step165/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c047a6448f39a1709f75ca908371d5c0d996437aeca11191907695094a1a5f59
+size 1047531685

output/finetuned_model/checkpoint-165/global_step165/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7c1b75d37d0366dd4bf6cb199a4881cb0c1437d6553b544235fef6d10dc147a
+size 1047531685

output/finetuned_model/checkpoint-165/global_step165/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e51cd9debf39609f3bab78aa077325c7d9a08d8d7d06a876ec9e5a04724a4818
+size 504721

output/finetuned_model/checkpoint-165/global_step165/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56eb5f5f4334596a4ebd8265059cd8e6c2b65727397430dd17166c34959b427c
+size 504721

output/finetuned_model/checkpoint-165/latest

@@ -0,0 +1 @@
+global_step165

output/finetuned_model/checkpoint-165/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e4105c99ac8eb33523066f111503ce7637a964f5a503fa603a8cdb2ea0be71a
+size 14917

output/finetuned_model/checkpoint-165/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84264b6e5bb199107adcbe258f1bd9ffc4e2a1e7e0f7097c3e8f46113aeeb092
+size 14917

output/finetuned_model/checkpoint-165/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4f4c7648800972a9eb6b3f5600e31f42ff22d228c98399ecf7b7b9711d81f46
+size 1465

output/finetuned_model/checkpoint-165/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}