Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: mistral-community/Mixtral-8x22B-v0.1
|
| 4 |
+
tags:
|
| 5 |
+
- generated_from_trainer
|
| 6 |
+
- axolotl
|
| 7 |
+
model-index:
|
| 8 |
+
- name: out
|
| 9 |
+
results: []
|
| 10 |
+
datasets:
|
| 11 |
+
- cognitivecomputations/Dolphin-2.9.2
|
| 12 |
+
- cognitivecomputations/SystemChat-2.0
|
| 13 |
+
- teknium/OpenHermes-2.5
|
| 14 |
+
- m-a-p/CodeFeedback-Filtered-Instruction
|
| 15 |
+
- cognitivecomputations/dolphin-coder
|
| 16 |
+
- cognitivecomputations/samantha-data
|
| 17 |
+
- HuggingFaceH4/ultrachat_200k
|
| 18 |
+
- microsoft/orca-math-word-problems-200k
|
| 19 |
+
- abacusai/SystemChat-1.1
|
| 20 |
+
- Locutusque/function-calling-chatml
|
| 21 |
+
- internlm/Agent-FLAN
|
| 22 |
+
language:
|
| 23 |
+
- en
|
| 24 |
+
---
|
| 25 |
+
|
| 26 |
+
# Dolphin 2.9.2 Mixtral 8x22b 🐬
|
| 27 |
+
|
| 28 |
+
Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations
|
| 29 |
+
|
| 30 |
+
[](https://discord.gg/cognitivecomputations)
|
| 31 |
+
Discord: https://discord.gg/cognitivecomputations
|
| 32 |
+
|
| 33 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/63111b2d88942700629f5771/ldkN1J0WIDQwU4vutGYiD.png" width="600" />
|
| 34 |
+
|
| 35 |
+
New in 2.9.2 is SystemChat 2.0 - a dataset designed to teach Dolphin to obey the system prompt, even over a long conversation.
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
My appreciation for the sponsors of Dolphin 2.9.2:
|
| 40 |
+
- [Crusoe Cloud](https://crusoe.ai/) - provided excellent on-demand 8xH100 node
|
| 41 |
+
- [OnDemand](https://on-demand.io/) - provided inference sponsorship, enabling creation of SystemChat
|
| 42 |
+
|
| 43 |
+
This model is based on Dolphin-2.9-Mixtral-8x22b, and is Apache-2.0 licensed.
|
| 44 |
+
|
| 45 |
+
The base model has 64k context, and fine-tuning was with 16k sequence length.
|
| 46 |
+
|
| 47 |
+
It took 1 week on 8xH100 provided by Crusoe Cloud
|
| 48 |
+
|
| 49 |
+
This model was trained FFT on 50% parameters (targeted with [Laser Scanner](https://github.com/cognitivecomputations/laserRMT/blob/main/laser_scanner.py) by Fernando Fernandes, David Golchinfar, Lucas Atkins, and Eric Hartford), using ChatML prompt template format.
|
| 50 |
+
|
| 51 |
+
example:
|
| 52 |
+
|
| 53 |
+
```
|
| 54 |
+
<|im_start|>system
|
| 55 |
+
You are Dolphin, a helpful AI assistant.<|im_end|>
|
| 56 |
+
<|im_start|>user
|
| 57 |
+
{prompt}<|im_end|>
|
| 58 |
+
<|im_start|>assistant
|
| 59 |
+
|
| 60 |
+
```
|
| 61 |
+
|
| 62 |
+
Dolphin-2.9 has a variety of instruction, conversational, and coding skills. It also has initial agentic abilities and supports function calling.
|
| 63 |
+
|
| 64 |
+
Dolphin is uncensored. I have filtered the dataset to remove alignment and bias. This makes the model more compliant. You are advised to implement your own alignment layer before exposing the model as a service. It will be highly compliant with any requests, even unethical ones. Please read my blog post about uncensored models. https://erichartford.com/uncensored-models You are responsible for any content you create using this model. Enjoy responsibly.
|
| 65 |
+
|
| 66 |
+
Dolphin is licensed Apache 2.0. I grant permission for any use, including commercial, that falls within accordance with Apache-2.0 license. Dolphin was trained on data generated from GPT4, among other models.
|
| 67 |
+
|
| 68 |
+
## Evals
|
| 69 |
+
|
| 70 |
+
TBD
|
| 71 |
+
|
| 72 |
+
## Training
|