#!/usr/bin/env bash
set -euo pipefail

  GET /console/api/workspaces/current/models/model-types/llm
  

props["max_tokens"] = 8192   # ❌ 错误位置

class ModelPropertyKey(StrEnum):
    MODE = auto()
    CONTEXT_SIZE = auto()
    MAX_CHUNKS = auto()
    FILE_UPLOAD_LIMIT = auto()
    SUPPORTED_FILE_EXTENSIONS = auto()
    MAX_CHARACTERS_PER_CHUNK = auto()
    DEFAULT_VOICE = auto()
    VOICES = auto()
    WORD_LIMIT = auto()
    AUDIO_TYPE = auto()
    MAX_WORKERS = auto()

/app/storage/cwd/langgenius/openai-1.0.3@<hash>/models/llm/llm.py

#!/usr/bin/env bash
set -euo pipefail

docker exec docker-plugin_daemon-1 sh -c 'find /app/storage/cwd -path "*langgenius/openai*" -name llm.py | head -1'

def get_customizable_model_schema(
        self, model: str, credentials: dict
    ) -> AIModelEntity:
    schema = {item.model: item for item in self.predefined_models()}.get(
        _base_model(model)
    )
    if schema is None:
        # Fallback for custom / non-OpenAI base models served via an
        # OpenAI-compatible endpoint (e.g. Qwen, DeepSeek, local vLLM):
        # clone a chat-capable predefined schema so the model is listable
        # and can be set as the default model.
        template = next(
            (m for m in self.predefined_models()
             if m.model_properties.get("mode") == "chat"),
            None,
        )
        if template is None:
            template = self.predefined_models()[0]
        props = dict(template.model_properties)
        mp = (credentials or {}).get("model_properties") or {}

        # Align context_size to the actual vLLM --max-model-len (65536) so Dify
        # never lets prompts exceed what the served model can accept. The
        # credential-update API does NOT persist model_properties for this
        # provider, so we hardcode the safe ceiling here as the default and
        # still honor an explicit override when one is supplied.
        VLLM_MAX_CTX = 64000
        try:
            props["context_size"] = int(mp["context_size"]) if "context_size" in mp else VLLM_MAX_CTX
        except (ValueError, TypeError):
            props["context_size"] = VLLM_MAX_CTX

        # NOTE: do NOT put max_tokens into model_properties.
        # Dify API-side ModelPropertyKey only allows {mode, context_size, ...}
        # and does NOT include MAX_TOKENS; adding it breaks Pydantic validation
        # when the API parses the schema from plugin_daemon.
        # max_tokens is already exposed as a parameter_rule by the cloned template.
        schema = template.model_copy(deep=True, update={"model_properties": props})
    return schema.model_copy(
        deep=True,
        update={
            "model": model,
            "label": I18nObject(zh_hans=model, en_us=model),
            "fetch_from": FetchFrom.CUSTOMIZABLE_MODEL,
        },
    )

# 只保留 context_size，不要写 max_tokens
cred["model_properties"] = {"context_size": NEW_CTX}

#!/usr/bin/env bash
set -euo pipefail

python3 -m py_compile llm.py && echo "SYNTAX OK"

#!/usr/bin/env bash
set -euo pipefail

# 假设已把修复后的 llm.py 放到 /tmp/llm_fixed.py
docker cp /tmp/llm_fixed.py docker-plugin_daemon-1:/app/storage/cwd/langgenius/openai-1.0.3@<hash>/models/llm/llm.py

# 重启插件守护进程
cd /dify/dify-main/docker   # 或 /dify/dify/docker
docker compose restart plugin_daemon

#!/usr/bin/env bash
set -euo pipefail

# 1) 确认 plugin_daemon 没有 SyntaxError / Traceback
docker logs docker-plugin_daemon-1 2>&1 | tail -30

# 2) 确认 API 不再报 ValueError
docker logs docker-api-1 2>&1 | grep -iE 'ValueError|model_schema|qwen3.6-27b' | tail -20

from dify_api import Dify
d = Dify("38205036@qq.com", "Difyai123456")
d.login()
r = d.get("/workspaces/current/models/model-types/llm")
print(r["text"])

{"data":[{"provider":"langgenius/openai/openai","models":[{"model":"qwen3.6-27b","model_properties":{"context_size":64000,"mode":"chat"},"status":"active"}]}]}

#!/usr/bin/env bash
set -euo pipefail

docker cp /tmp/llm_fixed.py docker-plugin_daemon-1:/app/storage/cwd/langgenius/openai-1.0.3@<hash>/models/llm/llm.py
cd /dify/dify-main/docker && docker compose restart plugin_daemon

#!/usr/bin/env bash
set -euo pipefail

# plugin_daemon 侧：确认 openai 插件正常加载
docker logs docker-plugin_daemon-1 2>&1 | tail -20

# API 侧：确认不再有 ValueError
docker logs docker-api-1 2>&1 | grep -iE 'ValueError|model_schema' | tail -20

#!/usr/bin/env bash
set -euo pipefail

# 用 dify_api.py 或 curl 调控制台 API
python - <<'PY'
from dify_api import Dify
d = Dify("38205036@qq.com", "Difyai123456")
d.login()
r = d.get("/workspaces/current/models/model-types/llm")
print(r["text"])
d.close()
PY

{"model":"qwen3.6-27b","model_properties":{"context_size":64000,"mode":"chat"},"status":"active"}

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""专题十 · 步骤 1：GET model-types/llm，确认 qwen3.6-27b 在列表中且无 schema 校验失败。"""
from __future__ import annotations

import json
import sys

import _bootstrap  # noqa: F401
import _config as C
from dify_api import Dify, pp


def main() -> int:
    d = Dify(C.DIFY_EMAIL, C.DIFY_PASSWORD)
    try:
        d.login()
        r = d.get("/workspaces/current/models/model-types/llm")
        pp("model-types/llm", r, limit=2000)
        data = (r.get("json") or {}).get("data", [])
        if not data:
            print("\n[FAIL] data 为空 —— 常见原因：model_properties 含非法 max_tokens")
            return 1
        found = False
        for p in data:
            if p.get("provider") != C.OPENAI_PROVIDER:
                continue
            for m in p.get("models", []):
                if m.get("model") == C.LLM_MODEL:
                    found = True
                    mp = m.get("model_properties") or {}
                    print("\n[OK] 找到模型:", json.dumps(m, ensure_ascii=False)[:500])
                    if "max_tokens" in mp:
                        print("[WARN] model_properties 仍含 max_tokens，可能再次触发校验失败")
        if not found:
            print("\n[FAIL] 未在列表中找到", C.LLM_MODEL)
            return 1
        return 0
    finally:
        d.close()


if __name__ == "__main__":
    sys.exit(main())

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""专题十 · 步骤 2 · 修改后：修正 llm.py，移除 model_properties 中的 max_tokens。"""
from __future__ import annotations

import sys

import _bootstrap  # noqa: F401
import _plugin_patch as P
from dify_api import Dify

import _config as C  # noqa: F401


def main() -> int:
    d = Dify(C.DIFY_EMAIL, C.DIFY_PASSWORD)
    try:
        d.login()
        llm_path = P.find_plugin_file(d, "llm.py")
        content = P.read_remote_file(d, llm_path)
        new_content = P.replace_method(
            content, "get_customizable_model_schema", P.GET_CUSTOMIZABLE_MODEL_SCHEMA
        )
        P.deploy_plugin_file(d, llm_path, new_content)
        print("已应用专题十 llm.py 补丁（仅 context_size，不含 max_tokens）")
        return 0
    finally:
        d.close()


if __name__ == "__main__":
    sys.exit(main())