Update model hub

[hahuyhoang411]

Motivation:
I want to choose best of the best models instead of 4-5 best model for each hardware range -> Optimize the plu in play style.

TODOs:

• Add base models (llama2 chat, mistral, yi) with some quantizations
• Decapriate models
• Make evaluation set to stretch test Discussion: Narrow down model list with best of the best models #827

GOAL:

• Update model.json
  • update models
  • update parameters

[hahuyhoang411]

Moved the HackMD evaluation questions here for ease https://hackmd.io/hhn9VFhGSD27_8LWMgAi6w

[tikikun]

Some current feedbacks on nightly model:

• Too many 33b and 70b, there models also need to be on the bottom of the list
• Too few 7-13b which is the most usable range for most system (for desktop app)
• Need to have "pinned" or equivalent in metadata to highlight best models, right now the first page of models are just models that is not desirable to use (1.5B 3B and 70b on first page)
• Naming of the model must be descriptive because we do per quantization per model, NeuralHermes 7b , but at which size?

[tikikun]

We need to add the below tags:
Recommended -> Small -> Medium -> Tiny - > Large

The tags are the order in which we need to follow through on the download page
Ref:
#839

[tikikun]

• Deepseek coder 1.3B not working for some reason, can you recheck once if this is an issue in nitro or it's frontend?
  https://github.com/janhq/jan/assets/22268502/aa36fd6a-b0d1-4097-b97d-1c0a7b56fdaa

• Need to replace OpenChat with a better version if there is

[hahuyhoang411]

On Nitro the deepseek 1.3b still works. I have no idea why it's not working on Jan.

Example result when using Deepseek 1.3b in Nitro

{
  "choices": [
    {
      "finish_reason": null,
      "index": 0,
      "message": {
        "content": "<jupyter_code>\n# Fibonacci series implementation using recursion\ndef fib(n):\n    if n <= 1:\n        return n\n    else:\n        return fib(n-1) + fib(n-2)\n\n# Driver code\nprint(fib(8))\n<jupyter_output>\n<empty_output>\n",
        "role": "assistant"
      }
    }
  ],
  "created": 1701687606,
  "id": "qeKmj6lya0rTugXDbfxt",
  "model": "_",
  "object": "chat.completion",
  "system_fingerprint": "_",
  "usage": {
    "completion_tokens": 86,
    "prompt_tokens": 81,
    "total_tokens": 167
  }
}

Image:

[louis-jan]

LGTM