biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[FP8] Extend per-token-group quantization support to QuantFP8 (#24342)

Browse Source 
Signed-off-by: Tahsin Tunan <tahsintunan@gmail.com>
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Luka Govedič <lgovedic@redhat.com>
This commit is contained in:
Tahsin Tunan
2025-09-17 07:31:06 +06:00
committed by GitHub
parent 493b10f8bf
commit cef32104b4
5 changed files with 444 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/layers/fused_moe/fused_moe.py
View File

@@ -32,9 +32,11 @@ from vllm.model_executor.layers.fused_moe.prepare_finalize import (
from vllm.model_executor.layers.fused_moe.topk_weight_and_reduce import (
TopKWeightAndReduceNoOP)
from vllm.model_executor.layers.fused_moe.utils import (
_resize_cache, moe_kernel_quantize_input, per_token_group_quant_fp8)
_resize_cache, moe_kernel_quantize_input)
from vllm.model_executor.layers.quantization.utils.flashinfer_utils import (
calculate_tile_tokens_dim)
from vllm.model_executor.layers.quantization.utils.fp8_utils import (
per_token_group_quant_fp8)
from vllm.model_executor.layers.quantization.utils.mxfp4_utils import (
dequant_mxfp4)
from vllm.platforms import current_platform