voxtral-mini-4b-asr-e3a-encfp8 / vllm_config.yaml
Shankara-A-S's picture
Upload folder using huggingface_hub
2f63cc2 verified
Raw
History Blame Contribute Delete
374 Bytes
served_model_name: voxtral-realtime
quantization: compressed-tensors
kv_cache_dtype: fp8_e4m3
attention_backend: TRITON_ATTN
tokenizer_mode: mistral
max_model_len: 16384
max_num_seqs: 1
max_num_batched_tokens: 16384
compilation_config:
cudagraph_mode: PIECEWISE
speculative_config:
method: ngram
num_speculative_tokens: 1
prompt_lookup_max: 2
prompt_lookup_min: 1