@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 30a2e80742
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 16, 09:30 UTC

[CI/Build] Add Model Tests for PixtralHF (#9813)

Michael Goin committed Nov 1, 2024 at 09:55 UTC 30a2e8074246e11a1452ab5e84a7be65ecac6119
1 file changed +9
tests/models/decoder_only/vision_language/test_models.py
+9
@@ -291,6 +291,15 @@ VLM_TEST_SETTINGS = {
291 # vllm_output_post_proc=model_utils.phi3v_vllm_to_hf_output,
292 # num_logprobs=10,
293 # ),
294 + "pixtral_hf": VLMTestInfo(
295 + models=["nm-testing/pixtral-12b-FP8-dynamic"],
296 + test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
297 + prompt_formatter=lambda img_prompt: f"<s>[INST]{img_prompt}[/INST]",
298 + img_idx_to_prompt=lambda idx: "[IMG]",
299 + max_model_len=8192,
300 + max_num_seqs=2,
301 + auto_cls=AutoModelForVision2Seq,
302 + ),
303 "qwen": VLMTestInfo(
304 models=["Qwen/Qwen-VL"],
305 test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),