Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Bagel-Hermes-2x34B
|
2 |
+
|
3 |
+
This is the model for Bagel-Hermes-2x34B. I used [mergekit](https://github.com/cg123/mergekit) to make this MOE model.
|
4 |
+
|
5 |
+
# Prompt Template(s):
|
6 |
+
|
7 |
+
Since [bagel-dpo-34b-v0.2](https://huggingface.co/jondurbin/bagel-dpo-34b-v0.2) uses many prompt templates, and [Nous-Hermes-2-Yi-34B](https://huggingface.co/NousResearch/Nous-Hermes-2-Yi-34B) uses ChatML, you can utilize ChatML and other prompt templates provided by bagel.
|
8 |
+
|
9 |
+
### ChatML:
|
10 |
+
|
11 |
+
```
|
12 |
+
<|im_start|>system
|
13 |
+
{system}<|im_end|>
|
14 |
+
<|im_start|>user
|
15 |
+
{user}<|im_end|>
|
16 |
+
<|im_start|>assistant
|
17 |
+
{asistant}<|im_end|>
|
18 |
+
```
|
19 |
+
|
20 |
+
### Alpaca (sort of)
|
21 |
+
|
22 |
+
```
|
23 |
+
Below is an instruction that describes a task. Write a response that appropriately completes the request.
|
24 |
+
|
25 |
+
### Instruction:
|
26 |
+
{system}
|
27 |
+
{instruction}
|
28 |
+
|
29 |
+
### Response:
|
30 |
+
```
|
31 |
+
### Vicuna
|
32 |
+
|
33 |
+
```
|
34 |
+
{system}
|
35 |
+
USER: {instruction}
|
36 |
+
ASSISTANT:
|
37 |
+
```
|
38 |
+
|
39 |
+
Visit [bagel-dpo-34b-v0.2](https://huggingface.co/jondurbin/bagel-dpo-34b-v0.2) to try more prompt templates.
|
40 |
+
|
41 |
+
# Yaml Config to reproduce
|
42 |
+
|
43 |
+
```yaml
|
44 |
+
base_model: nontoxic-bagel-34b-v0.2
|
45 |
+
gate_mode: hidden
|
46 |
+
dtype: bfloat16
|
47 |
+
|
48 |
+
experts:
|
49 |
+
- source_model: bagel-dpo-34b-v0.2
|
50 |
+
positive_prompts: ["question answering", "Q:", science", "biology", "chemistry", "physics"]
|
51 |
+
|
52 |
+
- source_model: Nous-Hermes-2-Yi-34B
|
53 |
+
positive_prompts: ["chat", "math", "reason", "mathematics", "solve", "count", "python", "javascript", "programming", "algorithm", "tell me", "assistant"]
|
54 |
+
```
|
55 |
+
|
56 |
+
# Quantizationed versions
|
57 |
+
|
58 |
+
Quantizationed versions of this model is available thanks to [TheBloke](https://hf.co/TheBloke).
|
59 |
+
|
60 |
+
##### GPTQ
|
61 |
+
|
62 |
+
- [TheBloke/Bagel-Hermes-2x34B-GPTQ](https://huggingface.co/TheBloke/Bagel-Hermes-2x34B-GPTQ)
|
63 |
+
|
64 |
+
##### GGUF
|
65 |
+
|
66 |
+
- [TheBloke/Bagel-Hermes-2x34B-GGUF](https://huggingface.co/TheBloke/Bagel-Hermes-2x34B-GGUF)
|
67 |
+
|
68 |
+
##### AWQ
|
69 |
+
|
70 |
+
- [TheBloke/Bagel-Hermes-2x34B-AWQ](https://huggingface.co/TheBloke/Bagel-Hermes-2x34B-AWQ)
|