Hugging Face's logo

RichardErkhov
/
chargoddard_-_llama-2-26b-trenchcoat-stack-gguf

GGUF
Inference Endpoints
Model card Files Community
RichardErkhov commited on
Commit
7eec58a
•
1 Parent(s): 0395e80

uploaded readme

Browse files
Files changed (1) hide show
  1. README.md +80 -0
README.md ADDED
@@ -0,0 +1,80 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Quantization made by Richard Erkhov.
2
+
3
+ [Github](https://github.com/RichardErkhov)
4
+
5
+ [Discord](https://discord.gg/pvy7H8DZMG)
6
+
7
+ [Request more models](https://github.com/RichardErkhov/quant_request)
8
+
9
+
10
+ llama-2-26b-trenchcoat-stack - GGUF
11
+ - Model creator: https://huggingface.co/chargoddard/
12
+ - Original model: https://huggingface.co/chargoddard/llama-2-26b-trenchcoat-stack/
13
+
14
+
15
+ | Name | Quant method | Size |
16
+ | ---- | ---- | ---- |
17
+ | [llama-2-26b-trenchcoat-stack.Q2_K.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q2_K.gguf) | Q2_K | 8.87GB |
18
+ | [llama-2-26b-trenchcoat-stack.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.IQ3_XS.gguf) | IQ3_XS | 9.8GB |
19
+ | [llama-2-26b-trenchcoat-stack.IQ3_S.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.IQ3_S.gguf) | IQ3_S | 10.35GB |
20
+ | [llama-2-26b-trenchcoat-stack.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q3_K_S.gguf) | Q3_K_S | 10.35GB |
21
+ | [llama-2-26b-trenchcoat-stack.IQ3_M.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.IQ3_M.gguf) | IQ3_M | 10.96GB |
22
+ | [llama-2-26b-trenchcoat-stack.Q3_K.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q3_K.gguf) | Q3_K | 11.62GB |
23
+ | [llama-2-26b-trenchcoat-stack.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q3_K_M.gguf) | Q3_K_M | 11.62GB |
24
+ | [llama-2-26b-trenchcoat-stack.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q3_K_L.gguf) | Q3_K_L | 12.72GB |
25
+ | [llama-2-26b-trenchcoat-stack.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.IQ4_XS.gguf) | IQ4_XS | 2.4GB |
26
+ | [llama-2-26b-trenchcoat-stack.Q4_0.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q4_0.gguf) | Q4_0 | 13.51GB |
27
+ | [llama-2-26b-trenchcoat-stack.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.IQ4_NL.gguf) | IQ4_NL | 13.59GB |
28
+ | [llama-2-26b-trenchcoat-stack.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q4_K_S.gguf) | Q4_K_S | 9.59GB |
29
+ | [llama-2-26b-trenchcoat-stack.Q4_K.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q4_K.gguf) | Q4_K | 14.44GB |
30
+ | [llama-2-26b-trenchcoat-stack.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q4_K_M.gguf) | Q4_K_M | 5.12GB |
31
+ | [llama-2-26b-trenchcoat-stack.Q4_1.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q4_1.gguf) | Q4_1 | 14.99GB |
32
+ | [llama-2-26b-trenchcoat-stack.Q5_0.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q5_0.gguf) | Q5_0 | 16.48GB |
33
+ | [llama-2-26b-trenchcoat-stack.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q5_K_S.gguf) | Q5_K_S | 16.48GB |
34
+ | [llama-2-26b-trenchcoat-stack.Q5_K.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q5_K.gguf) | Q5_K | 16.96GB |
35
+ | [llama-2-26b-trenchcoat-stack.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q5_K_M.gguf) | Q5_K_M | 16.96GB |
36
+ | [llama-2-26b-trenchcoat-stack.Q5_1.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q5_1.gguf) | Q5_1 | 17.97GB |
37
+ | [llama-2-26b-trenchcoat-stack.Q6_K.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q6_K.gguf) | Q6_K | 19.64GB |
38
+ | [llama-2-26b-trenchcoat-stack.Q8_0.gguf](https://huggingface.co/RichardErkhov/chargoddard_-_llama-2-26b-trenchcoat-stack-gguf/blob/main/llama-2-26b-trenchcoat-stack.Q8_0.gguf) | Q8_0 | 25.44GB |
39
+
40
+
41
+
42
+
43
+ Original model description:
44
+ ---
45
+ license: llama2
46
+ tags:
47
+ - merge
48
+ - mergekit
49
+ ---
50
+
51
+ Llama 2 13b is a pretty decent language model. You know what's probably better? *Two* Llama 2 13b models. In a trenchcoat.
52
+
53
+ Produced by [`bakllama.py`](https://github.com/cg123/mergekit/blob/main/bakllama.py) with this config file:
54
+ ```yml
55
+ layer_slices:
56
+ - model: TheBloke/Llama-2-13B-fp16
57
+ start: 0
58
+ end: 40
59
+ - model: TheBloke/Llama-2-13B-fp16
60
+ start: 0
61
+ end: 40
62
+ ```
63
+
64
+ No fine tuning was done on this model. Yes, it's still coherent somehow.
65
+
66
+ Benchmark results:
67
+ | Benchmark | Llama2-13b | Llama2-26b-tcs | Percent Change |
68
+ | --- | --- | --- | --- |
69
+ | ARC | 59.3 | 55.03 | -7.2% |
70
+ | HellaSwag | 82.15 | 79.9 | -2.74% |
71
+ | MMLU | 55.67 | 53.73| -3.48% |
72
+ | TruthfulQA | 37.39 | 40.48 | +5.59% |
73
+ | Average | 58.63 | 57.29 | -2.29% |
74
+ | Average Minus TQA | 65.70 | 62.85 | -4.34% |
75
+
76
+
77
+ This tells us two very important things:
78
+ 1. TruthfulQA is a perfect benchmark in every way.
79
+ 2. Llama models are amazingly robust to being fed their own output.
80
+