- Add first-time setup instructions for HF Jobs
- Include CLI installation and login steps
- Update HF Jobs documentation link to correct URL
- Add HF CLI guide link to resources
- Keep --secrets flag in all command examples
Update repository references to new uv-scripts org
- Changed all references from davanstrien/openai-oss to uv-scripts/openai-oss
- Updated HF Jobs commands in README
- Updated dataset card generation link
- Updated example commands in script docstring
- Now part of the uv-scripts organization
Add OpenAI GPT OSS collection link to documentation
- Added link to official OpenAI GPT OSS model collection
- Collection URL: https://huggingface.co/collections/openai/gpt-oss-68911959590a1634ba11c7a4
- Includes both 20B and 120B models in one place
- Added to README resources section
- Added to generated dataset cards for reference
- Creates minimal dataset card with generation details
- Documents dataset structure and usage
- Includes source dataset, model, and generation parameters
- Auto-pushed with dataset to HuggingFace Hub
Add temperature, top_p, and smart max_tokens defaults
- Set temperature=1.0 and top_p=1.0 as defaults (OpenAI recommended)
- Add --temperature and --top-p flags for customization
- Auto-scale max_new_tokens based on reasoning_effort:
- low: 512 tokens
- medium: 1024 tokens
- high: 2048 tokens (prevents truncation)
- Document OpenAI's sampling recommendations in README
- Users can still override with explicit --max-new-tokens
- Document tested and working l4x4 configuration
- Include exact command that worked
- Remove references to full script to avoid confusion
- Add clear GPU requirements table
- Explain reasoning_effort parameter
- Last tested: 2025-01-06
- Implements official reasoning_effort parameter (low/medium/high)
- Uses tokenizer.apply_chat_template with reasoning_effort
- Stores reasoning effort level in output dataset
- Updated to transformers>=4.55.0 for proper support
- Simplified version without channel parsing
- Stores raw outputs only
- Optimized for reliability on HF Jobs
- Works on regular GPUs (L4, A10G, A100)
- Based on official HF blog recommendations
Fix device_map for 120B model - use tp_plan instead
- Add gpt_oss_transformers.py with standard transformers library implementation
- Supports both 20B and 120B models with structured reasoning output
- Includes channel parsing for analysis (thinking) and final (response) channels
- Fixed device_map configuration for proper model loading
- Added MXFP4 quantization note (models are quantized out of the box)
- Comprehensive documentation with HF Jobs examples
This provides a fallback option while vLLM FlashAttention 3 support is pending.
