Text Generation
Transformers
PyTorch
English
experimental
research
bit-level
transformer
reversible
safety
telemetry
language-modeling
Instructions to use WCNegentropy/BitTransformerLM with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use WCNegentropy/BitTransformerLM with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="WCNegentropy/BitTransformerLM")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("WCNegentropy/BitTransformerLM", dtype="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps
- vLLM
How to use WCNegentropy/BitTransformerLM with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "WCNegentropy/BitTransformerLM" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "WCNegentropy/BitTransformerLM", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/WCNegentropy/BitTransformerLM
- SGLang
How to use WCNegentropy/BitTransformerLM with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "WCNegentropy/BitTransformerLM" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "WCNegentropy/BitTransformerLM", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "WCNegentropy/BitTransformerLM" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "WCNegentropy/BitTransformerLM", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use WCNegentropy/BitTransformerLM with Docker Model Runner:
docker model run hf.co/WCNegentropy/BitTransformerLM
π Add release notes for v2.0 production release
Browse files- RELEASE_NOTES.md +27 -0
RELEASE_NOTES.md
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# BitTransformerLM v2.0 - Production Release π
|
| 2 |
+
|
| 3 |
+
## Major Optimizations Implemented
|
| 4 |
+
|
| 5 |
+
β
**Performance Enhancements**
|
| 6 |
+
- Optimized run-length encoding with batch processing and parallel compression
|
| 7 |
+
- Memory-efficient chunked attention for long sequences with gradient checkpointing
|
| 8 |
+
- Advanced pipeline parallelism with load balancing and memory management
|
| 9 |
+
|
| 10 |
+
β
**Code Quality Improvements**
|
| 11 |
+
- Unified CLI flag naming conventions across all scripts
|
| 12 |
+
- Standardized function signatures with comprehensive type hints
|
| 13 |
+
- Comprehensive error recovery system with fallback mechanisms
|
| 14 |
+
|
| 15 |
+
β
**Production Readiness**
|
| 16 |
+
- Enhanced distributed training with FSDP and advanced communication optimization
|
| 17 |
+
- Robust error handling with graceful degradation
|
| 18 |
+
- Memory monitoring and automatic optimization
|
| 19 |
+
|
| 20 |
+
## Key Features
|
| 21 |
+
- **Bit-native Architecture**: Efficient processing of binary sequences
|
| 22 |
+
- **Safety Telemetry**: K/C/S metrics for model behavior monitoring
|
| 23 |
+
- **Reversible Layers**: Memory-efficient transformer architecture
|
| 24 |
+
- **Multi-format Support**: Run-length encoding, bit packing, diffusion mode
|
| 25 |
+
- **Distributed Training**: Advanced parallelism with automatic load balancing
|
| 26 |
+
|
| 27 |
+
Ready for production deployment and large-scale training workloads.
|