2026-03-05 17:27:17 +08:00
|
|
|
#
|
|
|
|
|
# Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
|
|
|
|
|
#
|
|
|
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
|
# you may not use this file except in compliance with the License.
|
|
|
|
|
# You may obtain a copy of the License at
|
|
|
|
|
#
|
|
|
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
|
#
|
|
|
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
|
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
|
# See the License for the specific language governing permissions and
|
|
|
|
|
# limitations under the License.
|
|
|
|
|
#
|
Fix: IMAGE2TEXT→CHAT fallback with model_type normalization in tenant_model_service (#14704)
## Summary
- When a model is registered as `chat` in `tenant_llm` but has the
`IMAGE2TEXT` tag in `llm_factories.json`, requesting it as `image2text`
(e.g. PDF parser) fails with `Tenant Model with name <model> and type
image2text not found`.
- After resolution via the new fallback, the returned
`config_dict["model_type"]` was still `"chat"`, causing
`tenant_llm_service.model_instance()` to instantiate `ChatModel` instead
of `CvModel` — breaking `describe_with_prompt` at ingestion time.
## What problem does this PR solve?
RAGFlow already has a `CHAT→IMAGE2TEXT` fallback: when a chat model is
not found, it retries with `image2text`. The symmetric fallback
(`IMAGE2TEXT→CHAT`) was missing.
This matters for multimodal models declared as `model_type: "chat"` with
an `IMAGE2TEXT` tag in `llm_factories.json` (e.g. models added after
tenant creation, or providers where a single model serves both
purposes). The frontend PDF parser selector correctly surfaces these
models via the `IMAGE2TEXT` tag, but the backend fails to resolve them
at runtime.
## Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
## Changes
**`api/db/joint_services/tenant_model_service.py`**
1. Add `IMAGE2TEXT→CHAT` fallback in
`get_model_config_by_type_and_name`: when an `image2text` model is not
found in `tenant_llm`, retry with `chat` — but only if the `llm` table
confirms `IMAGE2TEXT` capability via the `tags` field. This mirrors the
philosophy of the existing `CHAT→IMAGE2TEXT` fallback: substitution is
only allowed when the model has declared the required capability.
2. Normalize `config_dict["model_type"]` to `image2text` after the
fallback, so the caller (`model_instance`) correctly routes to `CvModel`
instead of `ChatModel`.
3. Extend the type validation guard to allow `(requested=image2text,
found=chat)` alongside the existing `(requested=chat, found=image2text)`
exception.
## Test plan
- [ ] Add a model with `model_type=chat` and `tags` containing
`IMAGE2TEXT` to a tenant
- [ ] Select it as PDF parser in a knowledge base
- [ ] Verify ingestion succeeds without `image2text not found` or
`describe_with_prompt` errors
- [ ] Verify the same model still works correctly in chat context
🤖 Generated with [Claude Code](https://claude.ai/claude-code)
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-09 04:40:58 +02:00
|
|
|
import logging
|
2026-03-05 17:27:17 +08:00
|
|
|
import os
|
|
|
|
|
import enum
|
2026-06-02 19:04:20 +08:00
|
|
|
import json
|
2026-03-05 17:27:17 +08:00
|
|
|
from common import settings
|
2026-06-10 14:59:57 +08:00
|
|
|
from common.constants import ActiveStatusEnum, LLMType, MINERU_DEFAULT_CONFIG, MINERU_ENV_KEYS, OPENDATALOADER_DEFAULT_CONFIG, OPENDATALOADER_ENV_KEYS, PADDLEOCR_DEFAULT_CONFIG, PADDLEOCR_ENV_KEYS
|
2026-06-10 13:04:13 +08:00
|
|
|
from api.db.services.tenant_llm_service import TenantService
|
2026-05-29 17:39:41 +08:00
|
|
|
from api.db.services.tenant_model_provider_service import TenantModelProviderService
|
|
|
|
|
from api.db.services.tenant_model_instance_service import TenantModelInstanceService
|
|
|
|
|
from api.db.services.tenant_model_service import TenantModelService
|
2026-03-05 17:27:17 +08:00
|
|
|
|
Fix: IMAGE2TEXT→CHAT fallback with model_type normalization in tenant_model_service (#14704)
## Summary
- When a model is registered as `chat` in `tenant_llm` but has the
`IMAGE2TEXT` tag in `llm_factories.json`, requesting it as `image2text`
(e.g. PDF parser) fails with `Tenant Model with name <model> and type
image2text not found`.
- After resolution via the new fallback, the returned
`config_dict["model_type"]` was still `"chat"`, causing
`tenant_llm_service.model_instance()` to instantiate `ChatModel` instead
of `CvModel` — breaking `describe_with_prompt` at ingestion time.
## What problem does this PR solve?
RAGFlow already has a `CHAT→IMAGE2TEXT` fallback: when a chat model is
not found, it retries with `image2text`. The symmetric fallback
(`IMAGE2TEXT→CHAT`) was missing.
This matters for multimodal models declared as `model_type: "chat"` with
an `IMAGE2TEXT` tag in `llm_factories.json` (e.g. models added after
tenant creation, or providers where a single model serves both
purposes). The frontend PDF parser selector correctly surfaces these
models via the `IMAGE2TEXT` tag, but the backend fails to resolve them
at runtime.
## Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
## Changes
**`api/db/joint_services/tenant_model_service.py`**
1. Add `IMAGE2TEXT→CHAT` fallback in
`get_model_config_by_type_and_name`: when an `image2text` model is not
found in `tenant_llm`, retry with `chat` — but only if the `llm` table
confirms `IMAGE2TEXT` capability via the `tags` field. This mirrors the
philosophy of the existing `CHAT→IMAGE2TEXT` fallback: substitution is
only allowed when the model has declared the required capability.
2. Normalize `config_dict["model_type"]` to `image2text` after the
fallback, so the caller (`model_instance`) correctly routes to `CvModel`
instead of `ChatModel`.
3. Extend the type validation guard to allow `(requested=image2text,
found=chat)` alongside the existing `(requested=chat, found=image2text)`
exception.
## Test plan
- [ ] Add a model with `model_type=chat` and `tags` containing
`IMAGE2TEXT` to a tenant
- [ ] Select it as PDF parser in a knowledge base
- [ ] Verify ingestion succeeds without `image2text not found` or
`describe_with_prompt` errors
- [ ] Verify the same model still works correctly in chat context
🤖 Generated with [Claude Code](https://claude.ai/claude-code)
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-09 04:40:58 +02:00
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
2026-03-05 17:27:17 +08:00
|
|
|
|
2026-06-10 15:35:21 +08:00
|
|
|
def _factory_model_types(llm: dict) -> list[str]:
|
|
|
|
|
model_type = llm.get("model_type")
|
|
|
|
|
if isinstance(model_type, list):
|
|
|
|
|
return model_type
|
|
|
|
|
return [model_type] if model_type else []
|
2026-06-10 13:04:13 +08:00
|
|
|
def _decode_api_key_config(raw_api_key: str) -> tuple[str, bool | None, str | None]:
|
|
|
|
|
if not raw_api_key:
|
|
|
|
|
return raw_api_key, None, None
|
|
|
|
|
|
|
|
|
|
try:
|
|
|
|
|
parsed = json.loads(raw_api_key)
|
|
|
|
|
except Exception:
|
|
|
|
|
return raw_api_key, None, None
|
|
|
|
|
|
|
|
|
|
if not isinstance(parsed, dict):
|
|
|
|
|
return raw_api_key, None, None
|
|
|
|
|
|
|
|
|
|
is_tools = bool(parsed["is_tools"]) if "is_tools" in parsed else None
|
|
|
|
|
if set(parsed.keys()) <= {"api_key", "is_tools"}:
|
|
|
|
|
return parsed.get("api_key", ""), is_tools, None
|
|
|
|
|
|
|
|
|
|
return parsed.get("api_key", raw_api_key), is_tools, raw_api_key
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def get_first_provider_model_name(tenant_id: str, provider_name: str, model_type: str | enum.Enum) -> str | None:
|
|
|
|
|
model_type_val = model_type if isinstance(model_type, str) else model_type.value
|
|
|
|
|
provider_obj = TenantModelProviderService.get_by_tenant_id_and_provider_name(tenant_id, provider_name)
|
|
|
|
|
if not provider_obj:
|
|
|
|
|
return None
|
|
|
|
|
|
|
|
|
|
for instance_obj in TenantModelInstanceService.get_all_by_provider_id(provider_obj.id):
|
|
|
|
|
if instance_obj.status != ActiveStatusEnum.ACTIVE.value:
|
|
|
|
|
continue
|
|
|
|
|
for model_obj in TenantModelService.get_models_by_instance_id(instance_obj.id):
|
|
|
|
|
if model_obj.model_type == model_type_val and model_obj.status == ActiveStatusEnum.ACTIVE.value:
|
|
|
|
|
return f"{model_obj.model_name}@{instance_obj.instance_name}@{provider_name}"
|
|
|
|
|
return None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _collect_env_config(env_keys: list[str], default_config: dict) -> dict | None:
|
|
|
|
|
config = dict(default_config)
|
|
|
|
|
found = False
|
|
|
|
|
for key in env_keys:
|
|
|
|
|
value = os.environ.get(key)
|
|
|
|
|
if value:
|
|
|
|
|
found = True
|
|
|
|
|
config[key] = value
|
|
|
|
|
return config if found else None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _ensure_ocr_provider_from_env(tenant_id: str, provider_name: str, model_name: str, config: dict | None) -> str | None:
|
|
|
|
|
if not config:
|
|
|
|
|
return None
|
|
|
|
|
|
|
|
|
|
provider_obj = TenantModelProviderService.get_by_tenant_id_and_provider_name(tenant_id, provider_name)
|
|
|
|
|
if not provider_obj:
|
|
|
|
|
TenantModelProviderService.insert(tenant_id=tenant_id, provider_name=provider_name)
|
|
|
|
|
provider_obj = TenantModelProviderService.get_by_tenant_id_and_provider_name(tenant_id, provider_name)
|
|
|
|
|
|
|
|
|
|
api_key = json.dumps(config)
|
|
|
|
|
instance_obj = TenantModelInstanceService.get_by_provider_id_and_api_key(provider_obj.id, api_key)
|
|
|
|
|
if not instance_obj:
|
|
|
|
|
instance_obj = TenantModelInstanceService.create_instance(
|
|
|
|
|
provider_id=provider_obj.id,
|
|
|
|
|
instance_name=model_name,
|
|
|
|
|
api_key=api_key,
|
|
|
|
|
extra="{}",
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
model_obj = TenantModelService.get_by_provider_id_and_instance_id_and_model_type_and_model_name(
|
|
|
|
|
provider_obj.id,
|
|
|
|
|
instance_obj.id,
|
|
|
|
|
LLMType.OCR.value,
|
|
|
|
|
model_name,
|
|
|
|
|
)
|
|
|
|
|
if not model_obj:
|
|
|
|
|
TenantModelService.insert(
|
|
|
|
|
model_name=model_name,
|
|
|
|
|
provider_id=provider_obj.id,
|
|
|
|
|
instance_id=instance_obj.id,
|
|
|
|
|
model_type=LLMType.OCR.value,
|
|
|
|
|
extra=json.dumps({"max_tokens": 0}),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
return f"{model_name}@{instance_obj.instance_name}@{provider_name}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def ensure_mineru_from_env(tenant_id: str) -> str | None:
|
|
|
|
|
return _ensure_ocr_provider_from_env(
|
|
|
|
|
tenant_id,
|
|
|
|
|
"MinerU",
|
|
|
|
|
"mineru-from-env",
|
|
|
|
|
_collect_env_config(MINERU_ENV_KEYS, MINERU_DEFAULT_CONFIG),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def ensure_paddleocr_from_env(tenant_id: str) -> str | None:
|
|
|
|
|
return _ensure_ocr_provider_from_env(
|
|
|
|
|
tenant_id,
|
|
|
|
|
"PaddleOCR",
|
|
|
|
|
"paddleocr-from-env",
|
|
|
|
|
_collect_env_config(PADDLEOCR_ENV_KEYS, PADDLEOCR_DEFAULT_CONFIG),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
2026-03-05 17:27:17 +08:00
|
|
|
def get_tenant_default_model_by_type(tenant_id: str, model_type: str|enum.Enum):
|
|
|
|
|
exist, tenant = TenantService.get_by_id(tenant_id)
|
|
|
|
|
if not exist:
|
|
|
|
|
raise LookupError("Tenant not found")
|
|
|
|
|
model_type_val = model_type if isinstance(model_type, str) else model_type.value
|
|
|
|
|
model_name: str = ""
|
|
|
|
|
match model_type_val:
|
|
|
|
|
case LLMType.EMBEDDING.value:
|
|
|
|
|
model_name = tenant.embd_id
|
|
|
|
|
case LLMType.SPEECH2TEXT.value:
|
|
|
|
|
model_name = tenant.asr_id
|
|
|
|
|
case LLMType.IMAGE2TEXT.value:
|
|
|
|
|
model_name = tenant.img2txt_id
|
|
|
|
|
case LLMType.CHAT.value:
|
|
|
|
|
model_name = tenant.llm_id
|
|
|
|
|
case LLMType.RERANK.value:
|
|
|
|
|
model_name = tenant.rerank_id
|
|
|
|
|
case LLMType.TTS.value:
|
|
|
|
|
model_name = tenant.tts_id
|
|
|
|
|
case LLMType.OCR.value:
|
|
|
|
|
raise Exception("OCR model name is required")
|
|
|
|
|
case _:
|
|
|
|
|
raise Exception(f"Unknown model type {model_type}")
|
|
|
|
|
if not model_name:
|
|
|
|
|
raise Exception(f"No default {model_type} model is set.")
|
2026-05-29 17:39:41 +08:00
|
|
|
return get_model_config_from_provider_instance(tenant_id, model_type, model_name)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def split_model_name(model_name: str):
|
|
|
|
|
# Parse model_name: {model_name} or {model_name}@{factory_name} or {model_name}@{instance_name}@{factory_name}
|
|
|
|
|
parts = model_name.split("@")
|
|
|
|
|
if len(parts) == 1:
|
|
|
|
|
pure_model_name = parts[0]
|
|
|
|
|
provider_name = ""
|
|
|
|
|
instance_name = ""
|
|
|
|
|
elif len(parts) == 2:
|
|
|
|
|
pure_model_name = parts[0]
|
|
|
|
|
provider_name = parts[1]
|
|
|
|
|
instance_name = "default"
|
|
|
|
|
else:
|
|
|
|
|
pure_model_name = parts[0]
|
|
|
|
|
instance_name = parts[1]
|
|
|
|
|
provider_name = parts[2]
|
|
|
|
|
return pure_model_name, instance_name, provider_name
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def get_model_config_from_provider_instance(tenant_id, model_type: str|enum.Enum, model_name: str):
|
|
|
|
|
pure_model_name, instance_name, provider_name = split_model_name(model_name)
|
|
|
|
|
model_type_val = model_type if isinstance(model_type, str) else model_type.value
|
|
|
|
|
# Builtin embedding model
|
|
|
|
|
compose_profiles = os.getenv("COMPOSE_PROFILES", "")
|
|
|
|
|
is_tei_builtin_embedding = (
|
|
|
|
|
model_type_val == LLMType.EMBEDDING.value
|
|
|
|
|
and "tei-" in compose_profiles
|
|
|
|
|
and pure_model_name == os.getenv("TEI_MODEL", "")
|
fix: show default embedding model when provider is not yet registered (#15511)
### What problem does this PR solve?
### Problem
On the Model Providers page, the Embedding Model dropdown in System
Model Settings shows empty (no default selected), even though a default
embedding model is configured in `service_conf.yaml`.
### Root Cause
Two issues were identified:
1. **Backend: `_get_model_info` fails for unregistered providers**
The tenant's `embd_id` is set to `bge-m3@xxxx` during initialization
(from the placeholder config `factory: 'xxxx'`). The `_get_model_info`
function requires the provider to exist in `tenant_model_provider`
table, but `xxxx` is never a real provider. Even after the user adds a
real provider (e.g., ZHIPU-AI), the stale `embd_id` still references the
non-existent one, causing the function to return `None`.
2. **Frontend: default models cache not invalidated after adding
provider**
`useAddProviderInstance` only invalidates `addedProviders` and
`allModels` caches after adding a provider instance, but does **not**
invalidate the `defaultModels` cache. This means the default model list
is not re-fetched until the user manually refreshes the page.
### Fix
**`api/apps/services/models_api_service.py`**
- Added `_resolve_model_from_tenant_providers()` helper: when the
default model's provider doesn't exist (e.g., placeholder `xxxx`), it
searches through the tenant's actually registered providers for a model
of the same type and returns the first match.
- When an instance name doesn't match (e.g., `"default"` vs actual name
`"1"`), the function now auto-resolves to the first real instance under
that provider.
- Falls back to `FACTORY_LLM_INFOS` validation when neither provider nor
instance exists.
**`web/src/hooks/use-llm-request.tsx`**
- Added `queryClient.invalidateQueries({ queryKey:
LlmKeys.defaultModels() })` to `useAddProviderInstance` so that the
default model list is re-fetched immediately after a provider instance
is added, eliminating the need for a manual page refresh.
### Testing
- Verified with a tenant whose `embd_id=bge-m3@xxxx` and only provider
is ZHIPU-AI (instance `1`): `_resolve_model_from_tenant_providers`
correctly resolves to `embedding-2@1@ZHIPU-AI`.
- After adding a provider via the UI, the embedding model dropdown now
immediately shows the resolved default without requiring a page refresh.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Signed-off-by: noob <yixiao121314@outlook.com>
2026-06-03 18:55:49 -07:00
|
|
|
and (provider_name == "Builtin" or not provider_name)
|
2026-05-29 17:39:41 +08:00
|
|
|
)
|
|
|
|
|
if is_tei_builtin_embedding:
|
|
|
|
|
# configured local embedding model
|
|
|
|
|
embedding_cfg = settings.EMBEDDING_CFG
|
|
|
|
|
return {
|
|
|
|
|
"llm_factory": "Builtin",
|
|
|
|
|
"api_key": embedding_cfg["api_key"],
|
|
|
|
|
"llm_name": pure_model_name,
|
|
|
|
|
"api_base": embedding_cfg["base_url"],
|
|
|
|
|
"model_type": LLMType.EMBEDDING.value,
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
provider_obj = TenantModelProviderService.get_by_tenant_id_and_provider_name(tenant_id, provider_name)
|
|
|
|
|
if not provider_obj:
|
|
|
|
|
raise LookupError(f"Provider {provider_name} not found for model {model_name}.")
|
|
|
|
|
instance_obj = TenantModelInstanceService.get_by_provider_id_and_instance_name(provider_obj.id, instance_name)
|
|
|
|
|
if not instance_obj:
|
|
|
|
|
raise LookupError(f"Instance {instance_name} not found for model {model_name}.")
|
|
|
|
|
model_obj = TenantModelService.get_by_provider_id_and_instance_id_and_model_type_and_model_name(provider_obj.id, instance_obj.id, model_type_val, pure_model_name)
|
|
|
|
|
|
2026-06-10 13:04:13 +08:00
|
|
|
api_key, is_tool, api_key_payload = _decode_api_key_config(instance_obj.api_key)
|
2026-05-29 17:39:41 +08:00
|
|
|
extra_fields = json.loads(instance_obj.extra) if instance_obj.extra else {}
|
|
|
|
|
|
|
|
|
|
if model_obj:
|
|
|
|
|
if model_obj.status == ActiveStatusEnum.INACTIVE.value:
|
|
|
|
|
raise LookupError(f"Model {model_name} is disabled.")
|
|
|
|
|
|
fix: propagate max_tokens from model config to downstream consumers (#15945)
## Summary
`get_model_config_from_provider_instance()` was not including
`max_tokens` in its returned dict, causing all downstream consumers
(dialog truncation, message fitting, knowledge base trimming, embedding,
graphrag, RAPTOR) to fall back to the hardcoded default of **8192
tokens** regardless of the actual model context window size (e.g.,
GPT-4o 128K, Claude 200K).
Closes #15944
## Root Cause
The function builds `model_config` with only: `llm_factory`, `api_key`,
`llm_name`, `api_base`, `model_type`, `is_tools`. `max_tokens` is never
included.
Yet the data exists in four independent sources:
1. `TenantModel.extra` JSON field — written by
`provider_api_service.py:659`
2. `conf/llm_factories.json` — every model entry has `max_tokens`
3. `rag/llm/model_meta.py` — 9 provider classes fetch real context
windows from APIs
4. `TenantLLM.max_tokens` database column
None of them are read by this function.
## Fix
Two lines added, one per return path:
- **Path B** (model_obj exists → provider-instance model): reads
`max_tokens` from `model_obj.extra` JSON
- **Path C** (fallback → factory config): reads `max_tokens` from
`llm_info` (sourced from `llm_factories.json`)
Both fall back to 8192 when the value is absent, preserving backward
compatibility.
## Impact
This single 5-line change fixes the context window budget for all **78+
call sites** across **20 files** that construct `LLMBundle` or read
`max_tokens` from the config dict, including:
| Consumer | File | Effect |
|---|---|---|
| Dialog chat truncation | `dialog_service.py:562` |
`message_fit_in(msg, max_tokens * 0.95)` now uses real context window |
| Knowledge base trimming | `dialog_service.py:752` |
`kb_prompt(kbinfos, max_tokens)` now fits more retrieved content |
| Agent message fitting | `agent/component/llm.py:322` | Agent prompts
no longer truncated at 7946 tokens |
| Embedding truncation | `task_executor.py:704` | Embedding input uses
actual model limit |
| GraphRAG extraction | `graphrag/*/extractor.py` | Entity extraction
gets full context budget |
| LLM4Tenant.max_length | `tenant_llm_service.py:513` | Chat model
wrapper exposes real context window |
2026-06-11 17:24:58 +08:00
|
|
|
model_extra = json.loads(model_obj.extra) if model_obj.extra else {}
|
2026-05-29 17:39:41 +08:00
|
|
|
model_config = {
|
|
|
|
|
"llm_factory": provider_obj.provider_name,
|
|
|
|
|
"api_key": api_key,
|
|
|
|
|
"llm_name": model_obj.model_name,
|
|
|
|
|
"api_base": extra_fields.get("base_url", ""),
|
|
|
|
|
"model_type": model_obj.model_type,
|
2026-06-11 17:29:28 +08:00
|
|
|
"is_tools": model_extra.get("is_tools", is_tool),
|
fix: propagate max_tokens from model config to downstream consumers (#15945)
## Summary
`get_model_config_from_provider_instance()` was not including
`max_tokens` in its returned dict, causing all downstream consumers
(dialog truncation, message fitting, knowledge base trimming, embedding,
graphrag, RAPTOR) to fall back to the hardcoded default of **8192
tokens** regardless of the actual model context window size (e.g.,
GPT-4o 128K, Claude 200K).
Closes #15944
## Root Cause
The function builds `model_config` with only: `llm_factory`, `api_key`,
`llm_name`, `api_base`, `model_type`, `is_tools`. `max_tokens` is never
included.
Yet the data exists in four independent sources:
1. `TenantModel.extra` JSON field — written by
`provider_api_service.py:659`
2. `conf/llm_factories.json` — every model entry has `max_tokens`
3. `rag/llm/model_meta.py` — 9 provider classes fetch real context
windows from APIs
4. `TenantLLM.max_tokens` database column
None of them are read by this function.
## Fix
Two lines added, one per return path:
- **Path B** (model_obj exists → provider-instance model): reads
`max_tokens` from `model_obj.extra` JSON
- **Path C** (fallback → factory config): reads `max_tokens` from
`llm_info` (sourced from `llm_factories.json`)
Both fall back to 8192 when the value is absent, preserving backward
compatibility.
## Impact
This single 5-line change fixes the context window budget for all **78+
call sites** across **20 files** that construct `LLMBundle` or read
`max_tokens` from the config dict, including:
| Consumer | File | Effect |
|---|---|---|
| Dialog chat truncation | `dialog_service.py:562` |
`message_fit_in(msg, max_tokens * 0.95)` now uses real context window |
| Knowledge base trimming | `dialog_service.py:752` |
`kb_prompt(kbinfos, max_tokens)` now fits more retrieved content |
| Agent message fitting | `agent/component/llm.py:322` | Agent prompts
no longer truncated at 7946 tokens |
| Embedding truncation | `task_executor.py:704` | Embedding input uses
actual model limit |
| GraphRAG extraction | `graphrag/*/extractor.py` | Entity extraction
gets full context budget |
| LLM4Tenant.max_length | `tenant_llm_service.py:513` | Chat model
wrapper exposes real context window |
2026-06-11 17:24:58 +08:00
|
|
|
"max_tokens": model_extra.get("max_tokens", 8192),
|
2026-05-29 17:39:41 +08:00
|
|
|
}
|
|
|
|
|
if api_key_payload is not None:
|
|
|
|
|
model_config["api_key_payload"] = api_key_payload
|
|
|
|
|
|
|
|
|
|
return model_config
|
|
|
|
|
else:
|
2026-06-02 19:04:20 +08:00
|
|
|
region = extra_fields.get("region", "default")
|
|
|
|
|
if region == "intl" and provider_name.lower() == "siliconflow":
|
|
|
|
|
target_factory_name = "siliconflow_intl"
|
|
|
|
|
else:
|
|
|
|
|
target_factory_name = provider_name
|
|
|
|
|
fac_list = [f for f in settings.FACTORY_LLM_INFOS if f["name"] == target_factory_name]
|
2026-05-29 17:39:41 +08:00
|
|
|
if not fac_list:
|
|
|
|
|
raise LookupError(f"Model provider config not found: {provider_name}")
|
|
|
|
|
llm_list = [llm for llm in fac_list[0]["llm"] if llm["llm_name"] == pure_model_name]
|
|
|
|
|
if not llm_list:
|
|
|
|
|
raise LookupError(f"Model config not found: {model_name}")
|
|
|
|
|
llm_info = llm_list[0]
|
2026-06-10 15:35:21 +08:00
|
|
|
if model_type_val not in _factory_model_types(llm_info):
|
|
|
|
|
raise LookupError(f"Model {model_name} is not a {model_type_val} model.")
|
2026-05-29 17:39:41 +08:00
|
|
|
model_config = {
|
|
|
|
|
"llm_factory": provider_obj.provider_name,
|
|
|
|
|
"api_key": api_key,
|
|
|
|
|
"llm_name": llm_info["llm_name"],
|
|
|
|
|
"api_base": extra_fields.get("base_url", ""),
|
2026-06-10 15:35:21 +08:00
|
|
|
"model_type": model_type_val,
|
fix: propagate max_tokens from model config to downstream consumers (#15945)
## Summary
`get_model_config_from_provider_instance()` was not including
`max_tokens` in its returned dict, causing all downstream consumers
(dialog truncation, message fitting, knowledge base trimming, embedding,
graphrag, RAPTOR) to fall back to the hardcoded default of **8192
tokens** regardless of the actual model context window size (e.g.,
GPT-4o 128K, Claude 200K).
Closes #15944
## Root Cause
The function builds `model_config` with only: `llm_factory`, `api_key`,
`llm_name`, `api_base`, `model_type`, `is_tools`. `max_tokens` is never
included.
Yet the data exists in four independent sources:
1. `TenantModel.extra` JSON field — written by
`provider_api_service.py:659`
2. `conf/llm_factories.json` — every model entry has `max_tokens`
3. `rag/llm/model_meta.py` — 9 provider classes fetch real context
windows from APIs
4. `TenantLLM.max_tokens` database column
None of them are read by this function.
## Fix
Two lines added, one per return path:
- **Path B** (model_obj exists → provider-instance model): reads
`max_tokens` from `model_obj.extra` JSON
- **Path C** (fallback → factory config): reads `max_tokens` from
`llm_info` (sourced from `llm_factories.json`)
Both fall back to 8192 when the value is absent, preserving backward
compatibility.
## Impact
This single 5-line change fixes the context window budget for all **78+
call sites** across **20 files** that construct `LLMBundle` or read
`max_tokens` from the config dict, including:
| Consumer | File | Effect |
|---|---|---|
| Dialog chat truncation | `dialog_service.py:562` |
`message_fit_in(msg, max_tokens * 0.95)` now uses real context window |
| Knowledge base trimming | `dialog_service.py:752` |
`kb_prompt(kbinfos, max_tokens)` now fits more retrieved content |
| Agent message fitting | `agent/component/llm.py:322` | Agent prompts
no longer truncated at 7946 tokens |
| Embedding truncation | `task_executor.py:704` | Embedding input uses
actual model limit |
| GraphRAG extraction | `graphrag/*/extractor.py` | Entity extraction
gets full context budget |
| LLM4Tenant.max_length | `tenant_llm_service.py:513` | Chat model
wrapper exposes real context window |
2026-06-11 17:24:58 +08:00
|
|
|
"is_tools": llm_info.get("is_tools", is_tool),
|
|
|
|
|
"max_tokens": llm_info.get("max_tokens", 8192),
|
2026-05-29 17:39:41 +08:00
|
|
|
}
|
|
|
|
|
if api_key_payload is not None:
|
|
|
|
|
model_config["api_key_payload"] = api_key_payload
|
|
|
|
|
return model_config
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def get_api_key(tenant_id: str, model_name: str):
|
|
|
|
|
_, instance_name, provider_name = split_model_name(model_name)
|
|
|
|
|
|
|
|
|
|
if not provider_name:
|
|
|
|
|
raise LookupError("Provider name is required.")
|
|
|
|
|
provider_obj = TenantModelProviderService.get_by_tenant_id_and_provider_name(tenant_id, provider_name)
|
|
|
|
|
if not provider_obj:
|
|
|
|
|
raise LookupError(f"Provider {provider_name} not found.")
|
|
|
|
|
instance_obj = TenantModelInstanceService.get_by_provider_id_and_instance_name(provider_obj.id, instance_name)
|
|
|
|
|
if not instance_obj:
|
|
|
|
|
raise LookupError(f"Instance {instance_name} not found.")
|
|
|
|
|
return instance_obj.api_key
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def get_model_type_by_name(tenant_id: str, model_name: str):
|
|
|
|
|
pure_model_name, instance_name, provider_name = split_model_name(model_name)
|
|
|
|
|
provider_obj = TenantModelProviderService.get_by_tenant_id_and_provider_name(tenant_id, provider_name)
|
|
|
|
|
if not provider_obj:
|
|
|
|
|
raise LookupError(f"Provider {provider_name} not found for model {model_name}.")
|
|
|
|
|
instance_obj = TenantModelInstanceService.get_by_provider_id_and_instance_name(provider_obj.id, instance_name)
|
|
|
|
|
if not instance_obj:
|
|
|
|
|
raise LookupError(f"Instance {instance_name} not found for model {model_name}.")
|
|
|
|
|
model_objs = TenantModelService.get_by_provider_id_and_instance_id_and_model_name(provider_obj.id, instance_obj.id, pure_model_name)
|
|
|
|
|
if not model_objs:
|
2026-06-02 19:04:20 +08:00
|
|
|
extra_fields = json.loads(instance_obj.extra) if instance_obj.extra else {}
|
|
|
|
|
region = extra_fields.get("region", "default")
|
|
|
|
|
if region == "intl" and provider_name.lower() == "siliconflow":
|
|
|
|
|
target_factory_name = "siliconflow_intl"
|
|
|
|
|
else:
|
|
|
|
|
target_factory_name = provider_name
|
|
|
|
|
fac_list = [f for f in settings.FACTORY_LLM_INFOS if f["name"] == target_factory_name]
|
2026-05-29 17:39:41 +08:00
|
|
|
if not fac_list:
|
|
|
|
|
raise LookupError(f"Model provider config not found: {provider_name}")
|
|
|
|
|
llm_list = [llm for llm in fac_list[0]["llm"] if llm["llm_name"] == pure_model_name]
|
|
|
|
|
if not llm_list:
|
|
|
|
|
raise LookupError(f"Model {pure_model_name} not found for model {model_name}.")
|
2026-06-10 15:35:21 +08:00
|
|
|
return _factory_model_types(llm_list[0])
|
2026-05-29 17:39:41 +08:00
|
|
|
return [model_obj.model_type for model_obj in model_objs]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def delete_models_by_instance_ids(instance_ids: list[str]):
|
|
|
|
|
return TenantModelService.delete_by_instance_ids(instance_ids)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def delete_instances_by_provider_ids(provider_ids: list[str]):
|
|
|
|
|
return TenantModelInstanceService.delete_by_provider_ids(provider_ids)
|
2026-06-10 14:59:57 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def ensure_opendataloader_from_env(tenant_id: str) -> str | None:
|
|
|
|
|
return _ensure_ocr_provider_from_env(
|
|
|
|
|
tenant_id,
|
|
|
|
|
"OpenDataLoader",
|
|
|
|
|
"opendataloader-from-env",
|
|
|
|
|
_collect_env_config(OPENDATALOADER_ENV_KEYS, OPENDATALOADER_DEFAULT_CONFIG),
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def get_models_by_tenant_and_provider_and_model_type(tenant_id: str, provider_name: str, model_type: str):
|
|
|
|
|
"""
|
|
|
|
|
Query TenantModel records by tenant_id, provider_name and model_name.
|
|
|
|
|
Returns all matching model records under all instances of the specified provider.
|
|
|
|
|
"""
|
|
|
|
|
provider_obj = TenantModelProviderService.get_by_tenant_id_and_provider_name(tenant_id, provider_name)
|
|
|
|
|
if not provider_obj:
|
|
|
|
|
return []
|
|
|
|
|
instances = TenantModelInstanceService.get_all_by_provider_id(provider_obj.id)
|
|
|
|
|
if not instances:
|
|
|
|
|
return []
|
|
|
|
|
results = []
|
|
|
|
|
for inst in instances:
|
|
|
|
|
models = TenantModelService.get_by_provider_id_and_instance_id_and_model_type(provider_obj.id, inst.id, model_type)
|
|
|
|
|
if models:
|
|
|
|
|
results.extend(models)
|
|
|
|
|
return results
|