Update README.md
Browse files
README.md
CHANGED
@@ -2,94 +2,4 @@
|
|
2 |
inference: true
|
3 |
---
|
4 |
|
5 |
-
|
6 |
-
https://huggingface.co/CRD716/ggml-vicuna-1.1-quantized/
|
7 |
-
|
8 |
-
---
|
9 |
-
|
10 |
-
### NOTE:
|
11 |
-
The PR [#1405](https://github.com/ggerganov/llama.cpp/pull/1405) brought breaking changes - none of the old models work with the latest build of llama.cpp.
|
12 |
-
|
13 |
-
Pre-PR #1405 files have been marked as old but remain accessible for those who need them.
|
14 |
-
|
15 |
-
Additionally, `q4_3` and `q4_2` have been completely axed in favor of their 5-bit counterparts (q5_1 and q5_0, respectively).
|
16 |
-
|
17 |
-
New files inference up to 10% faster without any quality reduction.
|
18 |
-
|
19 |
-
|
20 |
-
### Links
|
21 |
-
- [13B version of this model](https://huggingface.co/eachadea/ggml-vicuna-13b-1.1)
|
22 |
-
- [Set up with gpt4all-chat (one-click setup, available in in-app download menu)](https://gpt4all.io/index.html)
|
23 |
-
- [Set up with llama.cpp](https://github.com/ggerganov/llama.cpp)
|
24 |
-
- [Set up with oobabooga/text-generation-webui](https://github.com/oobabooga/text-generation-webui/blob/main/docs/llama.cpp-models.md)
|
25 |
-
|
26 |
-
### Info
|
27 |
-
- Main files are based on v1.1 release
|
28 |
-
- See changelog below
|
29 |
-
- Use prompt template: ```HUMAN: <prompt> ASSISTANT: <response>```
|
30 |
-
- Uncensored files are based on v0 release
|
31 |
-
- Use prompt template: ```### User: <prompt> ### Assistant: <response>```
|
32 |
-
- PR #896 was used for q4_0. Everything else is latest as of upload time.
|
33 |
-
|
34 |
-
### Quantization
|
35 |
-
Several quantization methods are supported. They differ in the resulting model disk size and inference speed.
|
36 |
-
|
37 |
-
Model | F16 | Q4_0 | Q4_1 | Q4_2 | Q4_3 | Q5_0 | Q5_1 | Q8_0
|
38 |
-
-- | -- | -- | -- | -- | -- | -- | -- | --
|
39 |
-
7B (ppl) | 5.9565 | 6.2103 | 6.1286 | 6.1698 | 6.0617 | 6.0139 | 5.9934 | 5.9571
|
40 |
-
7B (size) | 13.0G | 4.0G | 4.8G | 4.0G | 4.8G | 4.4G | 4.8G | 7.1G
|
41 |
-
7B (ms/tok @ 4th) | 128 | 56 | 61 | 84 | 91 | 91 | 95 | 75
|
42 |
-
7B (ms/tok @ 8th) | 128 | 47 | 55 | 48 | 53 | 53 | 59 | 75
|
43 |
-
7B (bpw) | 16.0 | 5.0 | 6.0 | 5.0 | 6.0 | 5.5 | 6.0 | 9.0
|
44 |
-
-- | -- | -- | -- | -- | -- | -- | -- | --
|
45 |
-
13B (ppl) | 5.2455 | 5.3748 | 5.3471 | 5.3433 | 5.3234 | 5.2768 | 5.2582 | 5.2458
|
46 |
-
13B (size) | 25.0G | 7.6G | 9.1G | 7.6G | 9.1G | 8.4G | 9.1G | 14G
|
47 |
-
13B (ms/tok @ 4th) | 239 | 104 | 113 | 160 | 175 | 176 | 185 | 141
|
48 |
-
13B (ms/tok @ 8th) | 240 | 85 | 99 | 97 | 114 | 108 | 117 | 147
|
49 |
-
13B (bpw) | 16.0 | 5.0 | 6.0 | 5.0 | 6.0 | 5.5 | 6.0 | 9.0
|
50 |
-
|
51 |
-
q5_1 or 5_0 are the latest and most performant implementations. The former is slightly more accurate at the cost of a bit of performance. Most users should use one of the two.
|
52 |
-
If you encounter any kind of compatibility issues, you might want to try the older q4_x
|
53 |
-
|
54 |
-
---
|
55 |
-
|
56 |
-
# Vicuna Model Card
|
57 |
-
|
58 |
-
## Model details
|
59 |
-
|
60 |
-
**Model type:**
|
61 |
-
Vicuna is an open-source chatbot trained by fine-tuning LLaMA on user-shared conversations collected from ShareGPT.
|
62 |
-
It is an auto-regressive language model, based on the transformer architecture.
|
63 |
-
|
64 |
-
**Model date:**
|
65 |
-
Vicuna was trained between March 2023 and April 2023.
|
66 |
-
|
67 |
-
**Organizations developing the model:**
|
68 |
-
The Vicuna team with members from UC Berkeley, CMU, Stanford, and UC San Diego.
|
69 |
-
|
70 |
-
**Paper or resources for more information:**
|
71 |
-
https://vicuna.lmsys.org/
|
72 |
-
|
73 |
-
**License:**
|
74 |
-
Apache License 2.0
|
75 |
-
|
76 |
-
**Where to send questions or comments about the model:**
|
77 |
-
https://github.com/lm-sys/FastChat/issues
|
78 |
-
|
79 |
-
## Intended use
|
80 |
-
**Primary intended uses:**
|
81 |
-
The primary use of Vicuna is research on large language models and chatbots.
|
82 |
-
|
83 |
-
**Primary intended users:**
|
84 |
-
The primary intended users of the model are researchers and hobbyists in natural language processing, machine learning, and artificial intelligence.
|
85 |
-
|
86 |
-
## Training dataset
|
87 |
-
70K conversations collected from ShareGPT.com.
|
88 |
-
(48k for the uncensored variant. 22k worth of garbage removed – see https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered)
|
89 |
-
|
90 |
-
## Evaluation dataset
|
91 |
-
A preliminary evaluation of the model quality is conducted by creating a set of 80 diverse questions and utilizing GPT-4 to judge the model outputs. See https://vicuna.lmsys.org/ for more details.
|
92 |
-
|
93 |
-
## Major updates of weights v1.1
|
94 |
-
- Refactor the tokenization and separator. In Vicuna v1.1, the separator has been changed from `"###"` to the EOS token `"</s>"`. This change makes it easier to determine the generation stop criteria and enables better compatibility with other libraries.
|
95 |
-
- Fix the supervised fine-tuning loss computation for better model quality.
|
|
|
2 |
inference: true
|
3 |
---
|
4 |
|
5 |
+
Obsolete model.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|