[FEAT]: Adding models for benchmarking

[clementou]

Description

TogetherAI models:

• [x] Yi-34B-Chat
• [x] Qwen-14B-Chat
• [x] WizardLM-13b-v1.2
• [x] Mistral-7b-Instruct-v0.2
• [x] deepseek-coder-33b-instruct
• [x] Vicuna v1.5 (13B)
• [x] CodeLlama-34B-Instruct
• [x] OpenHermes-2.5-Mistral-7B
• [x] openchat_3.5
• [x] alpaca (7B)
• [x] falcon-40b-instruct
• [x] StripedHyena-Nous-7B

Hugging Face models:

• [ ] Tulu-2-DP0-70B
• [ ] Dolphin-2.2.1-Mistral-7B
• [ ] Dolly-V2-12B
• [ ] deepseek-llm-67b-chat
• [ ] Zephyr-7b-beta
• [ ] Zephyr-7b-alpha
• [ ] FastChat-T5-3b
• [ ] WizardLM-70B-v1.0
• [ ] gpt4all-13b-snoozy
• [ ] oasst-sft-4-pythia-12b-epoch-3.5
• [ ] stablelm-tuned-alpha-7b
• [ ] Starling-LM-7B-alpha
• [ ] koala-13b-HF
• [ ] guanaco-33b-merged
• [ ] SOLAR-10.7B-Instruct-v.10

Tsinghua:

• [ ] chatglm4
• [ ] chatglm3-6b
• [ ] chatglm2-6b
• [ ] chatglm-6b

FastChat:

• [ ] Vicuna-33B

Nvidia:

• [ ] Llama2-70B-SteerLM-Chat

• [ ] rwkv-6-eagle

Proprietary:

Anthropic:

• [ ] Claude-1
• [ ] Claude-2.0
• [ ] Claude-2.1
• [ ] Claude-Instant-1

Google:

• [ ] gemini-pro
• [ ] PaLM-Chat-Bison-001

Mistral:

• [ ] Mistral Medium

Perplexity:

• [ ] pplx-70b-online
• [ ] pplx-7b-online

Additional Information

No response

[ProKil]

@clementou I feel like it is not scalable to add more models to LLM_Name. Since there are libraries like LiteLLM which converts all models to a uniform api, we can should allow using custom endpoints and not overflow our code with a long list of LLMs we support.

@lwaekfjlk @ruiyiw @Jasonqi146 @sharonwx54 worked on adding a custom model which could support all possible endpoints. There's no need to do repetitive efforts. Could y'all share how to use your custom_model option?

[XuhuiZhou]

Thanks @ProKil! @clementou Feel free to chat with people to figure out their solutions! You can directly post Qs in the sotopia-technical channel (or simply communicate here; i just find people usually don't check this haha)