[Bug Fix] Fix RunAI streamer: corrupted weights, missing quant init, and broken URIs for multimodal models (#22715)

Co-authored-by: Alex Nails <alex.nails@radixark.ai>

python/sglang/srt/utils/hf_transformers/common.py

@@ -52,6 +52,7 @@ from sglang.srt.configs import (
 from sglang.srt.configs.deepseek_ocr import DeepseekVLV2Config
 from sglang.srt.configs.internvl import InternVLChatConfig
 from sglang.srt.utils import get_bool_env_var, logger, lru_cache_frozenset
+from sglang.srt.utils.runai_utils import ObjectStorageModel, is_runai_obj_uri
 
 from ..hf_transformers_patches import normalize_rope_scaling_compat
 
@@ -140,6 +141,12 @@ def download_from_hf(
     return snapshot_download(model_path, allow_patterns=allow_patterns)
 
 
+def resolve_runai_obj_uri(model_name_or_path: str) -> str:
+    if is_runai_obj_uri(model_name_or_path):
+        return ObjectStorageModel.get_path(model_name_or_path)
+    return model_name_or_path
+
+
 def _resolve_local_or_cached_file(model_name_or_path, filename, revision=None):
     """Resolve a file from a local directory or HF hub cache (no network)."""
     local_path = Path(model_name_or_path) / filename

python/sglang/srt/utils/hf_transformers/config.py

@@ -20,7 +20,6 @@ from transformers.models.auto.modeling_auto import MODEL_FOR_CAUSAL_LM_MAPPING_N
 
 from sglang.srt.connector import create_remote_connector
 from sglang.srt.utils import is_remote_url, lru_cache_frozenset
-from sglang.srt.utils.runai_utils import ObjectStorageModel, is_runai_obj_uri
 
 from ..hf_transformers_patches import _ensure_gguf_version
 from .common import (
@@ -32,6 +31,7 @@ from .common import (
     _override_v_head_dim_if_zero,
     check_gguf_file,
     get_hf_text_config,
+    resolve_runai_obj_uri,
 )
 from .mistral_utils import is_mistral_model, load_mistral_config
 
@@ -60,8 +60,7 @@ def get_config(
         kwargs["gguf_file"] = model
         model = Path(model).parent
 
-    if is_runai_obj_uri(model):
+    model = resolve_runai_obj_uri(model)
-        model = ObjectStorageModel.get_path(model)
 
     if is_remote_url(model):
         client = create_remote_connector(model)

python/sglang/srt/utils/hf_transformers/processor.py

@@ -35,6 +35,7 @@ from .common import (
     attach_additional_stop_token_ids,
     download_from_hf,
     get_tokenizer_from_processor,
+    resolve_runai_obj_uri,
 )
 from .mistral_utils import (
     is_mistral_model,
@@ -150,6 +151,8 @@ def get_processor(
         _ensure_fastokens_patched()
 
     revision = kwargs.pop("revision", tokenizer_revision)
+    tokenizer_name = resolve_runai_obj_uri(tokenizer_name)
+
     if is_mistral_model(tokenizer_name):
         config = load_mistral_config(
             tokenizer_name,

python/sglang/srt/utils/hf_transformers/tokenizer.py

@@ -28,13 +28,13 @@ from transformers import (
 from sglang.srt.connector import create_remote_connector
 from sglang.srt.utils import is_remote_url, logger
 from sglang.srt.utils.patch_tokenizer import patch_tokenizer
-from sglang.srt.utils.runai_utils import ObjectStorageModel, is_runai_obj_uri
 
 from ..hf_transformers_patches import _ensure_gguf_version
 from .common import (
     _resolve_local_or_cached_file,
     attach_additional_stop_token_ids,
     check_gguf_file,
+    resolve_runai_obj_uri,
 )
 from .mistral_utils import (
     _MISTRAL_TOKENIZER_REDIRECTS,
@@ -146,8 +146,7 @@ def _resolve_tokenizer_name(tokenizer_name, kwargs):
         kwargs["gguf_file"] = tokenizer_name
         tokenizer_name = Path(tokenizer_name).parent
 
-    if is_runai_obj_uri(tokenizer_name):
+    tokenizer_name = resolve_runai_obj_uri(tokenizer_name)
-        tokenizer_name = ObjectStorageModel.get_path(tokenizer_name)
 
     if is_remote_url(tokenizer_name):
         # BaseConnector implements __del__() to clean up the local dir.