d8e949ec7c
feat: add Tavily search tab and integrate TavilyClient for web search functionality
1 month ago
41b7889a30
feat: integrate QA dataset loading and display gold answers in Gradio interface
1 month ago
7376f596a5
feat: add Gradio demo for DeepSearch and update configuration settings
1 month ago
7ff3623102
chore: update .gitignore, modify Makefile for installation, and add pyproject.toml for project configuration
1 month ago
eebf914a81
refactor: moved modules from src/deepsearch to src/
1 month ago
0f662d4330
refactor: moved FlashRAG submodule from src/ to third_party/
1 month ago
55f34b8503
feat: add FlashRAG as submodule
1 month ago
2fec4f2f42
refactor: change repo stucture (move code from src/ to src/deepsearch)
1 month ago
e3163081a0
docs: add experiment log for llama-3.2-3b-instruct experiments
1 month ago
010957cd99
feat: disable randomization option to get_qa_dataset function by default
1 month ago
56911a73f9
Update README.md
1 month ago
1a18cd7bfd
feat: update training and evaluation configurations (editable agent generation scripts)
...
Increased max_generations parameter in agentic_generate and run_eval functions for improved output flexibility.
1 month ago
77f121662f
test: add tests for reward_retry function scenarios
1 month ago
c8714e0f6b
feat: enhance reward_retry function to handle missing answer tags
...
Added logic to return 0 if the final message from the assistant does not contain answer tags (no matter how hard you try, you won't get anything if no result 💀 )
1 month ago
bf480574a2
fix: minor bug
1 month ago
3081d6e36b
test: added tests for new reward functions: search strategy and search diversity
1 month ago
4de31e0f30
feat: expand reward functions with new strategies and diversity checks
...
- Added reward functions for search strategy and search diversity
- Updated reward_format to include validation for proper message endings.
1 month ago
d0e6068055
fix: strengthen reward correctness logic to handle final message is not asnwer form assistant. Also update logs for reward functions for better debug
...
- Added 'logs/' directory to .gitignore to exclude log files.
- Introduced log_chat_state function to log chat states and rewards to JSONL files.
- Updated reward functions to log chat states with validation results for better tracking and debugging.
1 month ago
1bd609dfae
test: enhance reward correctness tests with validation logic
...
- Updated test cases to include role and tag validation for assistant messages.
- Ensured that only properly formatted messages with answer tags are accepted.
- Added new test for validating various incorrect formats and their expected outcomes.
1 month ago
338655e563
feat: refine user prompt logic for improved clarity and structure
1 month ago
6d994feeb2
feat: enhance evaluation scripts for base and LoRA models
1 month ago
da60b52bd1
feat: refactor download and upload scripts for improved argument handling (more notebook friendly :D)
1 month ago
fa3c0562fe
feat: add evaluation scripts for base and LoRA models
...
- Introduced `eval_base.py` for evaluating base model performance.
- Introduced `eval_lora.py` for evaluating LoRA model performance with additional LoRA weight handling.
1 month ago
1047e2fa1c
chore: update .gitignore and requirements for unsloth versions
1 month ago
83f86869f6
chore: update .gitignore and add new toys data files
1 month ago
133cb1ab90
test: add Qwen tokenizer adapter tests
...
Implemented unit tests for the Qwen tokenizer adapter, including format handling, mask generation, and multi-turn conversation support
1 month ago
6efe01d5ff
chore: update Makefile and requirements for testing
...
- Added a 'test' target in Makefile to run unit tests using pytest.
- Included 'wandb' in requirements.txt for experiment tracking.
1 month ago
af7f38c792
feat: add code for qwen architecture
1 month ago
e7915a6a8e
feat: add util script to upload/download checkpoints
1 month ago
9009440663
chore: disable logging, enable torch complie
1 month ago
d2f03b96ab
feat: enhance evaluation script and remove deprecated shell script
...
- Updated eval.py to streamline model evaluation using vLLM and unsloth.
- Deleted eval.sh as its functionality is now integrated into eval.py.
- Updated .gitignore to exclude eval_logs directory.
1 month ago
908768458c
chore: update Makefile and requirements for testing
...
- Added 'tests' directory to check_dirs in Makefile for better organization.
- Included 'pytest' in requirements.txt to facilitate unit testing.
1 month ago
90b45c62ab
docs: update docs and notebooks for the past few days, (observation, debugging)
...
- observation: model hallucniate the search result, docs about debugigng and adapting to r1 distil base model, notebooks on the detail of making training r1 distil works
1 month ago
3910ef343a
test: add unit tests for agent, reward functions, and tokenizer adapters
1 month ago
31dcbf5d8a
feat: refactor whole code base, add logic for training R1 distil base models, change some template and reward logics
...
- Break down rl_helpers into smaller modules
- Removed deprecated rl_helpers module to streamline the codebase.
- Enhance initial user prompt template inspired by Search-R1
1 month ago
c90c03267e
feat: change user prompt template to search-r1 inspried format
...
use <search></search> instead of embed whole tool definition, which resulted in lots or parsing errors
1 month ago
58dcf9a99d
refactor: simplify inference script by removing logger, load 16 bit model intead of raw lora finetuned
1 month ago
da79e986b6
feat: add new script and functionality in train script to save model in 16 bit format
1 month ago
f6b6cca2ce
feat: add multiple reference notebooks for model training and inference
...
Big thanks to author(s) for the great reference code!
1 month ago
04593fa8fd
style: change line length to 119, organize imports
1 month ago
abb18b10d8
feat: add CLI inference script with search functionality
...
This script is a bit dumb, but it worked. I'll update it later XD
1 month ago
fe70896023
chore: add Makefile for installation, code quality checks, style formatting, cleanup, and other tasks
1 month ago
60233f2113
chore: update .gitignore
1 month ago
fd32bcacfd
chores: update worklog and research progress
1 month ago
37730095a9
feat: add eval scripts that compare base model performance with the grpo trained model
1 month ago
7f2f43aa46
chore: clean up notebooks
1 month ago
3c2deaced9
refactor: restructure code base, better centralize logging logic
1 month ago
04d56325bb
feat: add new reward functions, add less dumb data generation logic, implement better logging
2 months ago
b22b02ea1d
feat: changed `<reasoning>` tags to `<think>
2 months ago
7d4de89186
chore: update worklog 250324
...
- Added `train_autodidact_1B.py` for quick test.
- Update `00_worklog.md`, `dataset.md`, and `reward-functions.md` to reflect new training strategies and reward functions.
2 months ago
thinhlpg
automaticcat