eab07f746c
* Add support for FP8 KV cache scales Since FP8 only has limited dynamic range, we can scale keys/values before storing them into the cache (and unscale them in attention). To avoid rescaling the cache as the absmax values change, good scales are usually determined per layer using calibration calibration data and stored in the checkpoint. This change adds support for for using key-value scales and loading them from checkpoints in the two most common formats: - Separate per-layer `k_scale` and `v_scale` scalars. - Per-layer `kv_scale` scalar (older format). Currently, scales are only used with an `float8_e4m3fn` cache. Besides adding support for key/value scales, the `fp8_quantize` function is also extended to support quantization with a kernel vendored from vLLM. This is slightly faster than the PyTorch implementation, but also scales in FP32, potentially improving accuracy. * Update FP8 KV cache test to use checkpoint with scales * `can_scale`: check that the attention is flashinfer |
||
|---|---|---|
| .. | ||
| custom_kernels | ||
| exllama_kernels | ||
| exllamav2_kernels | ||
| tests | ||
| text_generation_server | ||
| .gitignore | ||
| Makefile | ||
| Makefile-awq | ||
| Makefile-eetq | ||
| Makefile-exllamav2 | ||
| Makefile-fbgemm | ||
| Makefile-flash-att | ||
| Makefile-flash-att-v2 | ||
| Makefile-flashinfer | ||
| Makefile-lorax-punica | ||
| Makefile-selective-scan | ||
| Makefile-vllm | ||
| README.md | ||
| poetry.lock | ||
| pyproject.toml | ||
| requirements_cuda.txt | ||
| requirements_intel.txt | ||
| requirements_rocm.txt | ||
README.md
Text Generation Inference Python gRPC Server
A Python gRPC server for Text Generation Inference
Install
make install
Run
make run-dev