Upload generation/vllm_generator.py with huggingface_hub

generation/vllm_generator.py

@@ -1,11 +1,15 @@
 """VLLM-based text-to-speech generation logic with async streaming"""
 
 import asyncio
+import os
 import time
-import torch
+from pathlib import Path
+
 import numpy as np
-from vllm import AsyncEngineArgs, AsyncLLMEngine, SamplingParams
+import torch
+from huggingface_hub import list_repo_files, snapshot_download
 from transformers import AutoTokenizer
+from vllm import AsyncEngineArgs, AsyncLLMEngine, SamplingParams
 
 from config import (
     MODEL_NAME, TOKENIZER_NAME, START_OF_HUMAN, END_OF_TEXT, END_OF_HUMAN, END_OF_AI,
@@ -14,6 +18,68 @@ from config import (
 )
 
 
+def _pick_model_dir(candidate_dirs):
+    if not candidate_dirs:
+        return None
+    ranked_dirs = sorted(
+        {str(Path(directory)) for directory in candidate_dirs},
+        key=lambda directory: (
+            0 if "/models/" in directory else 1,
+            len(Path(directory).parts),
+            directory,
+        ),
+    )
+    return ranked_dirs[0]
+
+
+def _find_local_model_root(model_name):
+    model_path = Path(model_name).expanduser()
+    if not model_path.exists():
+        return None
+
+    if (model_path / "config.json").exists() or (model_path / "params.json").exists():
+        return str(model_path)
+
+    config_candidates = list(model_path.rglob("config.json")) + list(model_path.rglob("params.json"))
+    resolved_dir = _pick_model_dir(path.parent for path in config_candidates)
+    if resolved_dir:
+        print(f"Resolved local model root for vLLM: {resolved_dir}")
+    return resolved_dir
+
+
+def _resolve_model_name(model_name):
+    local_model_root = _find_local_model_root(model_name)
+    if local_model_root:
+        return local_model_root
+
+    try:
+        repo_files = list_repo_files(model_name)
+    except Exception:
+        return model_name
+
+    if "config.json" in repo_files or "params.json" in repo_files:
+        return model_name
+
+    config_candidates = [
+        file_path for file_path in repo_files
+        if file_path.endswith("/config.json") or file_path.endswith("/params.json")
+    ]
+    if not config_candidates:
+        return model_name
+
+    model_subdir = _pick_model_dir(Path(file_path).parent for file_path in config_candidates)
+    if model_subdir is None:
+        return model_name
+
+    snapshot_dir = snapshot_download(
+        repo_id=model_name,
+        allow_patterns=[f"{model_subdir}/*"],
+    )
+    resolved_model_root = os.path.join(snapshot_dir, model_subdir)
+    print(f"Resolved HF model root for vLLM: {resolved_model_root}")
+    return resolved_model_root
+
+
 class VLLMTTSGenerator:
     def __init__(self, tensor_parallel_size=1, gpu_memory_utilization=0.9, max_model_len=2048):
         """Initialize VLLM-based TTS generator with async streaming support
@@ -23,11 +89,12 @@ class VLLMTTSGenerator:
             gpu_memory_utilization: Fraction of GPU memory to use (0.0 to 1.0)
             max_model_len: Maximum sequence length
         """
-        print(f"Loading VLLM AsyncLLMEngine model: {MODEL_NAME}")
+        resolved_model_name = _resolve_model_name(MODEL_NAME)
+        print(f"Loading VLLM AsyncLLMEngine model: {resolved_model_name}")
 
         # Configure engine arguments
         engine_args = AsyncEngineArgs(
-            model=MODEL_NAME,
+            model=resolved_model_name,
             tokenizer=TOKENIZER_NAME,
             tensor_parallel_size=tensor_parallel_size,
             max_model_len=max_model_len,