[BUG] RD-Agent/LiteLLM: "LLM Provider NOT provided" even with correct provider/model for Ollama/DeepSeek (env, patch, CLI all fail)

[ericforai]

---

Description

When using the latest RD-Agent (0.6.x) and LiteLLM (1.72.x), it is impossible to use any non-OpenAI provider (e.g., Ollama, DeepSeek) for local LLM inference.
Even with all environment variables, CLI arguments, and code patches set to provider=ollama (or deepseek), RD-Agent always falls back to deepseek-chat and throws:

litellm.BadRequestError: LLM Provider NOT provided. Pass in the LLM provider you are trying to call. You passed model=deepseek-chat

This makes it impossible to use local LLMs or any non-OpenAI provider, even though the official documentation claims full support.

---

Environment

• OS: macOS 14.x (Apple Silicon)
• Python: 3.13.x
• RD-Agent: 0.6.0 / 0.6.1 (latest)
• LiteLLM: 1.72.4 (latest)
• Ollama: Installed and running (qwen2:7b model pulled, ollama serve active)
• DeepSeek API: Also tested, same error

---

Reproduction Steps

1. Set all environment variables for Ollama:

   export LITELLM_PROVIDER=ollama
   export LITELLM_MODEL=qwen2:7b
   export OLLAMA_BASE_URL=http://localhost:11434
   export RDAGENT_LLM_BACKEND=rdagent.oai.backend.litellm.LiteLLMBackend

2. (Also tried DeepSeek, same result)
3. Run:

   source rdagent_venv/bin/activate
   rdagent fin_factor --provider ollama --model qwen2:7b --max_iterations 3 --fast_mode

4. Result: Always fails with LLM Provider NOT provided, and falls back to deepseek-chat.

---

Error Log

litellm.BadRequestError: LLM Provider NOT provided. Pass in the LLM provider you are trying to call. You passed model=deepseek-chat
...
RuntimeError: Failed to create chat completion after 10 retries.

• Even after patching rdagent/oai/backend/litellm.py and base.py to forcibly inject provider="ollama" (or "deepseek"), the error persists.
• CLI arguments, environment variables, and code patching all fail to pass the provider to LiteLLM.

---

What I Tried

• Set all relevant environment variables (LITELLM_PROVIDER, LITELLM_MODEL, etc.)
• Used CLI arguments (--provider, --model)
• Patched rdagent/oai/backend/litellm.py and base.py to forcibly inject provider into kwargs
• Cleaned all .env files, restarted shell, reinstalled packages
• Confirmed Ollama is running and accessible
• Also tested with DeepSeek API (same error)

---

Expected Behavior

• RD-Agent should correctly pass the provider/model to LiteLLM and allow local LLM inference via Ollama (or DeepSeek API).
• The official documentation claims Ollama is supported, but the code does not work as described.

---

Additional Context

• This used to work in RD-Agent 0.5.x + LiteLLM 1.6x/1.7x (provider was not strictly required).
• The bug only appears after upgrading to RD-Agent 0.6.x and LiteLLM 1.72.x.
• There are similar issues reported in LiteLLM repo and SWE-agent.

---

Request

• Please provide a working example or fix for using Ollama (or any non-OpenAI provider) with RD-Agent 0.6.x + LiteLLM 1.72.x.
• If possible, clarify the correct way to pass provider/model through all layers (env, CLI, code) so that it is not overwritten or lost.

---

Thank you for your help!

---

如需补充具体 patch 代码或更详细的环境变量/命令行/报错堆栈，可随时补充。
建议直接复制此模板到 GitHub issue，官方会更容易定位和修复问题。

[Hoder-zyf]

maybe you can try our new version and as far as I know quite a lot people have succeeded in using deepseek model