Hugging Face's logo

v000000
/
NM-12B-Lyris-dev-2

Text Generation
Transformers
Safetensors
mistral
Merge
text-generation-inference
Inference Endpoints
Model card Files Community
NM-12B-Lyris-dev-2 / README.md
v000000's picture
v000000
Update README.md
6c7bd3e verified
preview code
|
raw
history blame
1.46 kB
metadata
base_model:
  - Sao10K/MN-12B-Lyra-v1
  - Sao10K/MN-12B-Lyra-v3
  - unsloth/Mistral-Nemo-Instruct-2407
library_name: transformers
tags:
  - merge
  - mistral
license: cc-by-nc-4.0

Lyris-dev2-Mistral-Nemo-12B-2407

image/png EXPERIMENTAL

attempt to fix Sao10k's Lyra-V3 prompt format and stop token >and boost smarts. with strategic LATCOS vector similarity merging

prototype, unfinished but works? Sometimes it does go on forever but it's way more useable, seems to have learnt to output stop token most of the time. But it's still pretty borked especially if greeting message is long. It needs even more Nemo-Instruct-2407 merged in.

  • Sao10K/MN-12B-Lyra-v1 Base
  • Sao10K/MN-12B-Lyra-v3 x2 Sequential PASS, order: 1, 3
  • unsloth/Mistral-Nemo-Instruct-2407 x1 Single PASS, order: 2
  • with z0.0001 value

Prompt format:

Mistral Instruct

[INST] System Message [/INST]

[INST] Name: Let's get started. Please respond based on the information and instructions provided above. [/INST]

<s>[INST] Name: What is your favourite condiment? [/INST]
AssistantName: Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!</s> 
[INST] Name: Do you have mayonnaise recipes? [/INST]