zlei6/ragflow
1
0
Fork 0
mirror of https://github.com/infiniflow/ragflow.git synced 2026-07-31 13:03:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

refa: resolve tenant model refs consistently (#16744)

Browse Source
This commit is contained in:
buua436
2026-07-09 14:02:08 +08:00
committed by GitHub
parent 794fcc2517
commit 6a77523bf0
51 changed files with 300 additions and 225 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
rag/flow/parser/parser.py
View File

@@ -32,7 +32,7 @@ from api.db.joint_services.tenant_model_service import (
ensure_opendataloader_from_env,
ensure_paddleocr_from_env,
get_first_provider_model_name,
get_model_config_from_provider_instance,
resolve_model_config,
get_tenant_default_model_by_type,
)
from common import settings
@@ -351,7 +351,7 @@ class Parser(ProcessBase):
elif lowered.endswith("@somark"):
# Keep the full 3-segment ``<llm_name>@<instance_name>@<provider>``
# form produced by the new Tenant LLM Provider UI (#14595);
# ``get_model_config_from_provider_instance`` -> ``split_model_name``
# ``resolve_model_config`` -> ``split_model_name``
# downstream requires all three segments.
parser_model_name = raw_parse_method
parse_method = "SoMark"
@@ -389,7 +389,7 @@ class Parser(ProcessBase):
raise RuntimeError("MinerU model not configured. Please add MinerU in Model Providers or set MINERU_* env.")
tenant_id = self._canvas._tenant_id
ocr_model_config = get_model_config_from_provider_instance(tenant_id, LLMType.OCR, parser_model_name)
ocr_model_config = resolve_model_config(tenant_id, LLMType.OCR, parser_model_name)
ocr_model = LLMBundle(tenant_id, ocr_model_config, lang=conf.get("lang", "Chinese"))
pdf_parser = ocr_model.mdl
@@ -462,7 +462,7 @@ class Parser(ProcessBase):
raise RuntimeError("OpenDataLoader model not configured. Please add OpenDataLoader in Model Providers.")
tenant_id = self._canvas._tenant_id
ocr_model_config = get_model_config_from_provider_instance(tenant_id, LLMType.OCR, parser_model_name)
ocr_model_config = resolve_model_config(tenant_id, LLMType.OCR, parser_model_name)
ocr_model = LLMBundle(tenant_id, ocr_model_config)
pdf_parser = ocr_model.mdl
@@ -524,7 +524,7 @@ class Parser(ProcessBase):
tenant_id = self._canvas._tenant_id
try:
ocr_model_config = get_model_config_from_provider_instance(tenant_id, LLMType.OCR, parser_model_name)
ocr_model_config = resolve_model_config(tenant_id, LLMType.OCR, parser_model_name)
except Exception:
if "@" in parser_model_name:
raise
@@ -614,7 +614,7 @@ class Parser(ProcessBase):
raise RuntimeError("PaddleOCR model not configured. Please add PaddleOCR in Model Providers or set PADDLEOCR_* env.")
tenant_id = self._canvas._tenant_id
ocr_model_config = get_model_config_from_provider_instance(tenant_id, LLMType.OCR, parser_model_name)
ocr_model_config = resolve_model_config(tenant_id, LLMType.OCR, parser_model_name)
ocr_model = LLMBundle(tenant_id, ocr_model_config)
pdf_parser = ocr_model.mdl
@@ -644,7 +644,7 @@ class Parser(ProcessBase):
# Vision parser treats each page as a large image block.
else:
if conf.get("parse_method"):
vision_model_config = get_model_config_from_provider_instance(self._canvas._tenant_id, LLMType.IMAGE2TEXT, conf["parse_method"])
vision_model_config = resolve_model_config(self._canvas._tenant_id, LLMType.IMAGE2TEXT, conf["parse_method"])
else:
vision_model_config = get_tenant_default_model_by_type(self._canvas._tenant_id, LLMType.IMAGE2TEXT)
vision_model = LLMBundle(self._canvas._tenant_id, vision_model_config, lang=self._param.setups["pdf"].get("lang"))
@@ -1117,7 +1117,7 @@ class Parser(ProcessBase):
else:
lang = conf["lang"]
# use VLM to describe the picture
cv_model_config = get_model_config_from_provider_instance(self._canvas.get_tenant_id(), LLMType.IMAGE2TEXT, conf["parse_method"])
cv_model_config = resolve_model_config(self._canvas.get_tenant_id(), LLMType.IMAGE2TEXT, conf["parse_method"])
cv_model = LLMBundle(self._canvas.get_tenant_id(), cv_model_config, lang=lang)
img_binary = io.BytesIO()
img.save(img_binary, format="JPEG")
@@ -1153,7 +1153,7 @@ class Parser(ProcessBase):
tmpf.write(blob)
tmpf.flush()
tmp_path = os.path.abspath(tmpf.name)
seq2txt_model_config = get_model_config_from_provider_instance(self._canvas.get_tenant_id(), LLMType.SPEECH2TEXT, vlm["llm_id"])
seq2txt_model_config = resolve_model_config(self._canvas.get_tenant_id(), LLMType.SPEECH2TEXT, vlm["llm_id"])
seq2txt_mdl = LLMBundle(self._canvas.get_tenant_id(), seq2txt_model_config)
txt = seq2txt_mdl.transcription(tmp_path)
@@ -1166,7 +1166,7 @@ class Parser(ProcessBase):
conf = self._param.setups["video"]
vlm = conf.get("vlm")
self.set_output("output_format", conf["output_format"])
cv_model_config = get_model_config_from_provider_instance(self._canvas.get_tenant_id(), LLMType.IMAGE2TEXT, vlm["llm_id"])
cv_model_config = resolve_model_config(self._canvas.get_tenant_id(), LLMType.IMAGE2TEXT, vlm["llm_id"])
cv_mdl = LLMBundle(self._canvas.get_tenant_id(), cv_model_config)
video_prompt = str(conf.get("prompt", "") or "")
txt = asyncio.run(cv_mdl.async_chat(system="", history=[], gen_conf={}, video_bytes=blob, filename=name, video_prompt=video_prompt))