I think I found the key to avoid repetitions and long rambling answers, which this model has a tendency to do. Hopefully a further fine tune will reduce it. The key is to reduce creativity all the way down, and make the model deterministic. How do you do that?, you may ask. Easy, it is controlled by the following 3 inference parameters: temp, top_p, and top_k

With the following default settings I often get repetitions or additional rambling information:

    "top_k": 40,
    "top_p": 0.95,
    "temp": 0.8,

If I use the following values instead, to make the model deterministic, the problem seems to be gone:

    "top_k": 1,
    "top_p": 0.1,
    "temp": 0.1,

Please note that if you want to use the model for story writing, maybe you get better results by dialing up the creativity.

Here is my complete config file for LM Studio:

{
  "name": "OpenChat",
  "inference_params": {
    "top_k": 1,
    "top_p": 0.1,
    "temp": 0.1,
    "input_prefix": "GPT4 Correct User: ",
    "input_suffix": "<|end_of_turn|>GPT4 Correct Assistant: ",
    "antiprompt": [
      "GPT4",
      "<|end_of_turn|>",
      "[End of Turn]",
      "[]"
    ],
    "pre_prompt": "Below is an instruction that describes a task. Write a concise response that appropriately completes the request. Ensure all essential details are provided. Each of your statements must be unique.",
    "pre_prompt_suffix": "<|end_of_turn|>",
    "pre_prompt_prefix": "GPT4 System: "
  }
}

A few words about the above:

• I only include necessary options to avoid overwriting user settings when loading the model or switching prompt format. If you export a config file, please make sure you then edit it manually to clean it up.
• GPT Correct User/Assistant. The Correct keyword is important. It refers to the training data, where the answers were verified as correct. If you do not use it (eg: GPT4 User), it will still works, but it will five more weight to training data which was unverified (Human User was also used)
• GPT4 Sytem or just System are the 2 official recommended ways to prefix system messages. Either work.
• In my system message (pre_promt), I avoid any negative (eg: No repetitions). Remember this is just a language model: if it sees the word “repeat” (or similar), it will have a tendency to see this as an instruction to create repetitions! Instead I turned it around into a positive statement based on the word “unique”.
• Trailing spaces in the prefixes and suffixes are not critical, but ensure proper formatting.

​

As a bonus, here is my config for generating code, which according to my limited testing, this model seems to be surprisingly good at:

{
  "name": "OpenChat Code",
  "inference_params": {
    "top_k": 1,
    "top_p": 0.1,
    "temp": 0.1,
    "input_prefix": "Code User: ",
    "input_suffix": "<|end_of_turn|>Code Assistant: ",
    "antiprompt": [
      "GPT4",
      "<|end_of_turn|>",
      "[End of Turn]",
      "[]"
    ],
    "pre_prompt": "You are a helpful coding assistant. Respond concisely, but ensure all essential details are provided. Each of your statements must be unique.",
    "pre_prompt_suffix": "<|end_of_turn|>",
    "pre_prompt_prefix": "GPT4 System: "
  }
}

not sure about that, but it is good.

How was the model size increased to 11B. It’s a merge but with what?

I got starling-11b-q8_0.gguf in LM studio and can’t get a decent output. Or I get a one liner like “here is a possible response: hello” with defaut or chatml preset, or i get pages of various “possible” responses filed with smileys most of the time (alpaca preset).

What chat format can I use?

I saw their 7B model closing in on gpt-4 scores in some benchmarks which is absolutely wild but also sus

with itself.

you can now download it with lmstudio. be sure to use the openchat prompt template.

gguf files have finished uploading for all llama.cpp users.

mergekit. there is a link in the post.

I’m just testing it casually rn in lmsys, and really like it’s flow and tone. It’s pretty pleasant to speak to from the get go, it’s a good start and I can’t wait to dig a bit harder on it.

Great model. Question, do you know why it’s outputting these “<0x0A>” tokens?

For example:

Here are some ways to improve your vertical leap:**<0x0A><0x0A>**1. Strength training: Focus on exercises

I tried this model a little while ago with LM Studio and I noticed that it does not have GPU acceleration. Sin

I uploaded my lmstudio config file to https://huggingface.co/NurtureAI/Starling-LM-11B-alpha-v1-GGUF to help people with lmstudio setup.

also i released more 11b mistral sizes… incase anyone is interested.

https://huggingface.co/NurtureAI/SynthIA-11B-v1.3

https://huggingface.co/NurtureAI/Mistral-11B-Instruct-v0.1

https://huggingface.co/NurtureAI/dolphin-2.2.1-mistral-11b

https://huggingface.co/NurtureAI/zephyr-11b-beta

https://huggingface.co/NurtureAI/openchat_3.5-11B
https://huggingface.co/NurtureAI/neural-chat-11b-v3-1

May the force be with you.

what are ur settings for passthrough as im trying to make a 11b https://huggingface.co/maywell/Synatra-7B-v0.3-RP