Upload 13 files
Browse files- .gitattributes +12 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ1_M.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ1_S.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ2_M.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ2_S.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ2_XS.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ2_XXS.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ3_M.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ3_S.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ3_XS.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ3_XXS.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL.gguf +3 -0
- DeepSeek-Coder-V2-Lite-Instruct.imatrix.dat +3 -0
- README.md +425 -5
.gitattributes
CHANGED
@@ -33,3 +33,15 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
DeepSeek-Coder-V2-Lite-Instruct.imatrix.dat filter=lfs diff=lfs merge=lfs -text
|
37 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ1_M.gguf filter=lfs diff=lfs merge=lfs -text
|
38 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ1_S.gguf filter=lfs diff=lfs merge=lfs -text
|
39 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ2_M.gguf filter=lfs diff=lfs merge=lfs -text
|
40 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ2_S.gguf filter=lfs diff=lfs merge=lfs -text
|
41 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ2_XS.gguf filter=lfs diff=lfs merge=lfs -text
|
42 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ2_XXS.gguf filter=lfs diff=lfs merge=lfs -text
|
43 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ3_M.gguf filter=lfs diff=lfs merge=lfs -text
|
44 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ3_S.gguf filter=lfs diff=lfs merge=lfs -text
|
45 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ3_XS.gguf filter=lfs diff=lfs merge=lfs -text
|
46 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ3_XXS.gguf filter=lfs diff=lfs merge=lfs -text
|
47 |
+
DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL.gguf filter=lfs diff=lfs merge=lfs -text
|
DeepSeek-Coder-V2-Lite-Instruct.IQ1_M.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2db3987649c0c306e33e677c2e763990dec3cb0f90645dd8f4620c670ac5659c
|
3 |
+
size 5236564832
|
DeepSeek-Coder-V2-Lite-Instruct.IQ1_S.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:54b1ec4b3275209203ea0515e9d27cc1d1ebfb05f214702bc6ec7e1a9c80d76f
|
3 |
+
size 4994131808
|
DeepSeek-Coder-V2-Lite-Instruct.IQ2_M.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e01dafb76ff2f6e780f23a6e58f401b84cc100b28c21f1f4d12d5de470203e0b
|
3 |
+
size 6328457056
|
DeepSeek-Coder-V2-Lite-Instruct.IQ2_S.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1a5d61a56206b00f131291f76e00970c3a1573bcc5f25852783fbb9dcfb10c97
|
3 |
+
size 6005213024
|
DeepSeek-Coder-V2-Lite-Instruct.IQ2_XS.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:51898f3d6f17d88d8b6a7de0bbab926c29b4871222937ae18957d101b5cf9285
|
3 |
+
size 5967402848
|
DeepSeek-Coder-V2-Lite-Instruct.IQ2_XXS.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ded772822a1a745587b3c0a43a1a2919d2ba4dfc98b5f679166ccd7c83a4ebbf
|
3 |
+
size 5640619872
|
DeepSeek-Coder-V2-Lite-Instruct.IQ3_M.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:5819e9b5e5b4ca4a125f0488049f4956b54fb7852d65433fd70999d15c2b93e0
|
3 |
+
size 7553175392
|
DeepSeek-Coder-V2-Lite-Instruct.IQ3_S.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:daf8aa5e71d59e3610a855cb27b5a90a66b1c250faefdc21ae61819cc61d5966
|
3 |
+
size 7487663968
|
DeepSeek-Coder-V2-Lite-Instruct.IQ3_XS.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:fa74fd31b9c0fdf615be9d9304c95bfe86b54e1791944ab8a60f9d0b8fc48ca5
|
3 |
+
size 7122857824
|
DeepSeek-Coder-V2-Lite-Instruct.IQ3_XXS.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:bd716cb24a9d4e978430ca27ee67395de68a3766c37b681f609ef4bcf28a4f48
|
3 |
+
size 6964057952
|
DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL.gguf
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:be9b3d5b29c561b38d6c41e81c0be12244bf6a4cb1d2b9a45175d3339ae040f4
|
3 |
+
size 8905110368
|
DeepSeek-Coder-V2-Lite-Instruct.imatrix.dat
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b7853ddc8576c89f434a99960b5eff2718284b599b09c10b166bb17469a77c47
|
3 |
+
size 38356411
|
README.md
CHANGED
@@ -1,5 +1,425 @@
|
|
1 |
-
---
|
2 |
-
license: other
|
3 |
-
license_name: deepseek-license
|
4 |
-
license_link: https://github.com/deepseek-ai/DeepSeek-Coder-V2/raw/main/LICENSE-MODEL
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: other
|
3 |
+
license_name: deepseek-license
|
4 |
+
license_link: https://github.com/deepseek-ai/DeepSeek-Coder-V2/raw/main/LICENSE-MODEL
|
5 |
+
tags:
|
6 |
+
- code
|
7 |
+
language:
|
8 |
+
- code
|
9 |
+
base_model: deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
|
10 |
+
model_creator: DeepSeek AI
|
11 |
+
model_name: DeepSeek-Coder-V2-Lite-Instruct
|
12 |
+
model_type: deepseek2
|
13 |
+
datasets:
|
14 |
+
- m-a-p/CodeFeedback-Filtered-Instruction
|
15 |
+
quantized_by: CISC
|
16 |
+
---
|
17 |
+
|
18 |
+
# DeepSeek-Coder-V2-Lite-Instruct - SOTA GGUF
|
19 |
+
- Model creator: [DeepSeek AI](https://huggingface.co/deepseek-ai)
|
20 |
+
- Original model: [DeepSeek-Coder-V2-Lite-Instruct](https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct)
|
21 |
+
|
22 |
+
<!-- description start -->
|
23 |
+
## Description
|
24 |
+
|
25 |
+
This repo contains State Of The Art quantized GGUF format model files for [DeepSeek-Coder-V2-Lite-Instruct](https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct).
|
26 |
+
|
27 |
+
Quantization was done with an importance matrix that was trained for ~250K tokens (64 batches of 4096 tokens) of answers from the [CodeFeedback-Filtered-Instruction](https://huggingface.co/datasets/m-a-p/CodeFeedback-Filtered-Instruction) dataset.
|
28 |
+
|
29 |
+
Fill-in-Middle token metadata has been added, see [example](#simple-llama-cpp-python-example-fill-in-middle-code).
|
30 |
+
|
31 |
+
NOTE: Due to some of the tensors in this model being oddly shaped a consequential portion of the quantization fell back to IQ4_NL instead of the specified method, causing somewhat larger (and "smarter"; even IQ1_M is quite usable) model files than usual!
|
32 |
+
|
33 |
+
<!-- description end -->
|
34 |
+
|
35 |
+
|
36 |
+
<!-- prompt-template start -->
|
37 |
+
## Prompt template: DeepSeek v2
|
38 |
+
|
39 |
+
```
|
40 |
+
User: {prompt}
|
41 |
+
|
42 |
+
Assistant:
|
43 |
+
```
|
44 |
+
|
45 |
+
<!-- prompt-template end -->
|
46 |
+
|
47 |
+
|
48 |
+
<!-- compatibility_gguf start -->
|
49 |
+
## Compatibility
|
50 |
+
|
51 |
+
These quantised GGUFv3 files are compatible with llama.cpp from May 29th 2024 onwards, as of commit [fb76ec2](https://github.com/ggerganov/llama.cpp/commit/fb76ec31a9914b7761c1727303ab30380fd4f05c)
|
52 |
+
|
53 |
+
They are also compatible with many third party UIs and libraries provided they are built using a recent llama.cpp.
|
54 |
+
|
55 |
+
## Explanation of quantisation methods
|
56 |
+
|
57 |
+
<details>
|
58 |
+
<summary>Click to see details</summary>
|
59 |
+
|
60 |
+
The new methods available are:
|
61 |
+
|
62 |
+
* GGML_TYPE_IQ1_S - 1-bit quantization in super-blocks with an importance matrix applied, effectively using 1.56 bits per weight (bpw)
|
63 |
+
* GGML_TYPE_IQ1_M - 1-bit quantization in super-blocks with an importance matrix applied, effectively using 1.75 bpw
|
64 |
+
* GGML_TYPE_IQ2_XXS - 2-bit quantization in super-blocks with an importance matrix applied, effectively using 2.06 bpw
|
65 |
+
* GGML_TYPE_IQ2_XS - 2-bit quantization in super-blocks with an importance matrix applied, effectively using 2.31 bpw
|
66 |
+
* GGML_TYPE_IQ2_S - 2-bit quantization in super-blocks with an importance matrix applied, effectively using 2.5 bpw
|
67 |
+
* GGML_TYPE_IQ2_M - 2-bit quantization in super-blocks with an importance matrix applied, effectively using 2.7 bpw
|
68 |
+
* GGML_TYPE_IQ3_XXS - 3-bit quantization in super-blocks with an importance matrix applied, effectively using 3.06 bpw
|
69 |
+
* GGML_TYPE_IQ3_XS - 3-bit quantization in super-blocks with an importance matrix applied, effectively using 3.3 bpw
|
70 |
+
* GGML_TYPE_IQ3_S - 3-bit quantization in super-blocks with an importance matrix applied, effectively using 3.44 bpw
|
71 |
+
* GGML_TYPE_IQ3_M - 3-bit quantization in super-blocks with an importance matrix applied, effectively using 3.66 bpw
|
72 |
+
* GGML_TYPE_IQ4_XS - 4-bit quantization in super-blocks with an importance matrix applied, effectively using 4.25 bpw
|
73 |
+
* GGML_TYPE_IQ4_NL - 4-bit non-linearly mapped quantization with an importance matrix applied, effectively using 4.5 bpw
|
74 |
+
|
75 |
+
Refer to the Provided Files table below to see what files use which methods, and how.
|
76 |
+
</details>
|
77 |
+
<!-- compatibility_gguf end -->
|
78 |
+
|
79 |
+
<!-- README_GGUF.md-provided-files start -->
|
80 |
+
## Provided files
|
81 |
+
|
82 |
+
| Name | Quant method | Bits | Size | Max RAM required | Use case |
|
83 |
+
| ---- | ---- | ---- | ---- | ---- | ----- |
|
84 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ1_S.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ1_S.gguf) | IQ1_S | 1 | 4.5 GB| 5.5 GB | smallest, significant quality loss |
|
85 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ1_M.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ1_M.gguf) | IQ1_M | 1 | 4.7 GB| 5.7 GB | very small, significant quality loss |
|
86 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ2_XXS.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ2_XXS.gguf) | IQ2_XXS | 2 | 5.1 GB| 6.1 GB | very small, high quality loss |
|
87 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ2_XS.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ2_XS.gguf) | IQ2_XS | 2 | 5.4 GB| 6.4 GB | very small, high quality loss |
|
88 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ2_S.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ2_S.gguf) | IQ2_S | 2 | 5.4 GB| 6.4 GB | small, substantial quality loss |
|
89 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ2_M.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ2_M.gguf) | IQ2_M | 2 | 5.7 GB| 5.7 GB | small, greater quality loss |
|
90 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ3_XXS.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ3_XXS.gguf) | IQ3_XXS | 3 | 6.3 GB| 7.3 GB | very small, high quality loss |
|
91 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ3_XS.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ3_XS.gguf) | IQ3_XS | 3 | 6.5 GB| 7.5 GB | small, substantial quality loss |
|
92 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ3_S.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ3_S.gguf) | IQ3_S | 3 | 6.8 GB| 7.8 GB | small, greater quality loss |
|
93 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ3_M.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ3_M.gguf) | IQ3_M | 3 | 6.9 GB| 7.9 GB | medium, balanced quality - recommended |
|
94 |
+
| [DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL.gguf](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL.gguf) | IQ4_NL | 4 | 8.1 GB| 9.1 GB | small, substantial quality loss |
|
95 |
+
|
96 |
+
Generated importance matrix file: [DeepSeek-Coder-V2-Lite-Instruct.imatrix.dat](https://huggingface.co/CISCai/DeepSeek-Coder-V2-Lite-Instruct-SOTA-GGUF/blob/main/DeepSeek-Coder-V2-Lite-Instruct.imatrix.dat)
|
97 |
+
|
98 |
+
**Note**: the above RAM figures assume no GPU offloading with 4K context. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
|
99 |
+
|
100 |
+
<!-- README_GGUF.md-provided-files end -->
|
101 |
+
|
102 |
+
<!-- README_GGUF.md-how-to-run start -->
|
103 |
+
## Example `llama.cpp` command
|
104 |
+
|
105 |
+
Make sure you are using `llama.cpp` from commit [fb76ec3](https://github.com/ggerganov/llama.cpp/commit/fb76ec31a9914b7761c1727303ab30380fd4f05c) or later.
|
106 |
+
|
107 |
+
```shell
|
108 |
+
./llama-cli -ngl 28 -m DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL.gguf --color -c 131072 --temp 0 --repeat-penalty 1.1 -p "User: {prompt}\n\nAssistant:"
|
109 |
+
```
|
110 |
+
|
111 |
+
Change `-ngl 28` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
|
112 |
+
|
113 |
+
Change `-c 131072` to the desired sequence length.
|
114 |
+
|
115 |
+
If you are low on V/RAM try quantizing the K-cache with `-ctk q8_0` or even `-ctk q4_0` for big memory savings (depending on context size).
|
116 |
+
There is a similar option for V-cache (`-ctv`), however that requires Flash Attention [which is not working yet with this model](https://github.com/ggerganov/llama.cpp/issues/7343).
|
117 |
+
|
118 |
+
For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md)
|
119 |
+
|
120 |
+
## How to run from Python code
|
121 |
+
|
122 |
+
You can use GGUF models from Python using the [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) module.
|
123 |
+
|
124 |
+
### How to load this model in Python code, using llama-cpp-python
|
125 |
+
|
126 |
+
For full documentation, please see: [llama-cpp-python docs](https://llama-cpp-python.readthedocs.io/en/latest/).
|
127 |
+
|
128 |
+
#### First install the package
|
129 |
+
|
130 |
+
Run one of the following commands, according to your system:
|
131 |
+
|
132 |
+
```shell
|
133 |
+
# Prebuilt wheel with basic CPU support
|
134 |
+
pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu
|
135 |
+
# Prebuilt wheel with NVidia CUDA acceleration
|
136 |
+
pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu121 (or cu122 etc.)
|
137 |
+
# Prebuilt wheel with Metal GPU acceleration
|
138 |
+
pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/metal
|
139 |
+
# Build base version with no GPU acceleration
|
140 |
+
pip install llama-cpp-python
|
141 |
+
# With NVidia CUDA acceleration
|
142 |
+
CMAKE_ARGS="-DLLAMA_CUDA=on" pip install llama-cpp-python
|
143 |
+
# Or with OpenBLAS acceleration
|
144 |
+
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
|
145 |
+
# Or with CLBLast acceleration
|
146 |
+
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
|
147 |
+
# Or with AMD ROCm GPU acceleration (Linux only)
|
148 |
+
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
|
149 |
+
# Or with Metal GPU acceleration for macOS systems only
|
150 |
+
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
|
151 |
+
# Or with Vulkan acceleration
|
152 |
+
CMAKE_ARGS="-DLLAMA_VULKAN=on" pip install llama-cpp-python
|
153 |
+
# Or with Kompute acceleration
|
154 |
+
CMAKE_ARGS="-DLLAMA_KOMPUTE=on" pip install llama-cpp-python
|
155 |
+
# Or with SYCL acceleration
|
156 |
+
CMAKE_ARGS="-DLLAMA_SYCL=on -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx" pip install llama-cpp-python
|
157 |
+
|
158 |
+
# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
|
159 |
+
$env:CMAKE_ARGS = "-DLLAMA_CUDA=on"
|
160 |
+
pip install llama-cpp-python
|
161 |
+
```
|
162 |
+
|
163 |
+
#### Simple llama-cpp-python example code
|
164 |
+
|
165 |
+
```python
|
166 |
+
from llama_cpp import Llama
|
167 |
+
|
168 |
+
# Chat Completion API
|
169 |
+
|
170 |
+
llm = Llama(model_path="./DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL.gguf", n_gpu_layers=28, n_ctx=131072)
|
171 |
+
print(llm.create_chat_completion(
|
172 |
+
repeat_penalty = 1.1,
|
173 |
+
messages = [
|
174 |
+
{
|
175 |
+
"role": "user",
|
176 |
+
"content": "Pick a LeetCode challenge and solve it in Python."
|
177 |
+
}
|
178 |
+
]
|
179 |
+
))
|
180 |
+
```
|
181 |
+
|
182 |
+
#### Simple llama-cpp-python example fill-in-middle code
|
183 |
+
|
184 |
+
```python
|
185 |
+
from llama_cpp import Llama
|
186 |
+
|
187 |
+
# Completion API
|
188 |
+
|
189 |
+
prompt = "def add("
|
190 |
+
suffix = "\n return sum\n\n"
|
191 |
+
|
192 |
+
llm = Llama(model_path="./DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL.gguf", n_gpu_layers=28, n_ctx=131072)
|
193 |
+
output = llm.create_completion(
|
194 |
+
temperature = 0.0,
|
195 |
+
repeat_penalty = 1.0,
|
196 |
+
prompt = prompt,
|
197 |
+
suffix = suffix
|
198 |
+
)
|
199 |
+
|
200 |
+
# Models sometimes repeat suffix in response, attempt to filter that
|
201 |
+
response = output["choices"][0]["text"]
|
202 |
+
response_stripped = response.rstrip()
|
203 |
+
unwanted_response_suffix = suffix.rstrip()
|
204 |
+
unwanted_response_length = len(unwanted_response_suffix)
|
205 |
+
|
206 |
+
filtered = False
|
207 |
+
if unwanted_response_suffix and response_stripped[-unwanted_response_length:] == unwanted_response_suffix:
|
208 |
+
response = response_stripped[:-unwanted_response_length]
|
209 |
+
filtered = True
|
210 |
+
|
211 |
+
print(f"Fill-in-Middle completion{' (filtered)' if filtered else ''}:\n\n{prompt}\033[32m{response}\033[{'33' if filtered else '0'}m{suffix}\033[0m")
|
212 |
+
```
|
213 |
+
|
214 |
+
<!-- README_GGUF.md-how-to-run end -->
|
215 |
+
|
216 |
+
<!-- original-model-card start -->
|
217 |
+
<!-- markdownlint-disable first-line-h1 -->
|
218 |
+
<!-- markdownlint-disable html -->
|
219 |
+
<!-- markdownlint-disable no-duplicate-header -->
|
220 |
+
|
221 |
+
<div align="center">
|
222 |
+
<img src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.svg?raw=true" width="60%" alt="DeepSeek-V2" />
|
223 |
+
</div>
|
224 |
+
<hr>
|
225 |
+
<div align="center" style="line-height: 1;">
|
226 |
+
<a href="https://www.deepseek.com/" target="_blank" style="margin: 2px;">
|
227 |
+
<img alt="Homepage" src="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/badge.svg?raw=true" style="display: inline-block; vertical-align: middle;"/>
|
228 |
+
</a>
|
229 |
+
<a href="https://chat.deepseek.com/" target="_blank" style="margin: 2px;">
|
230 |
+
<img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-DeepSeek%20V2-536af5?color=536af5&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
|
231 |
+
</a>
|
232 |
+
<a href="https://huggingface.co/deepseek-ai" target="_blank" style="margin: 2px;">
|
233 |
+
<img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-DeepSeek%20AI-ffc107?color=ffc107&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
|
234 |
+
</a>
|
235 |
+
</div>
|
236 |
+
|
237 |
+
<div align="center" style="line-height: 1;">
|
238 |
+
<a href="https://discord.gg/Tc7c45Zzu5" target="_blank" style="margin: 2px;">
|
239 |
+
<img alt="Discord" src="https://img.shields.io/badge/Discord-DeepSeek%20AI-7289da?logo=discord&logoColor=white&color=7289da" style="display: inline-block; vertical-align: middle;"/>
|
240 |
+
</a>
|
241 |
+
<a href="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/figures/qr.jpeg?raw=true" target="_blank" style="margin: 2px;">
|
242 |
+
<img alt="Wechat" src="https://img.shields.io/badge/WeChat-DeepSeek%20AI-brightgreen?logo=wechat&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
|
243 |
+
</a>
|
244 |
+
<a href="https://twitter.com/deepseek_ai" target="_blank" style="margin: 2px;">
|
245 |
+
<img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-deepseek_ai-white?logo=x&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
|
246 |
+
</a>
|
247 |
+
</div>
|
248 |
+
|
249 |
+
<div align="center" style="line-height: 1;">
|
250 |
+
<a href="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-CODE" style="margin: 2px;">
|
251 |
+
<img alt="Code License" src="https://img.shields.io/badge/Code_License-MIT-f5de53?&color=f5de53" style="display: inline-block; vertical-align: middle;"/>
|
252 |
+
</a>
|
253 |
+
<a href="https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL" style="margin: 2px;">
|
254 |
+
<img alt="Model License" src="https://img.shields.io/badge/Model_License-Model_Agreement-f5de53?&color=f5de53" style="display: inline-block; vertical-align: middle;"/>
|
255 |
+
</a>
|
256 |
+
</div>
|
257 |
+
<p align="center">
|
258 |
+
<a href="#4-api-platform">API Platform</a> |
|
259 |
+
<a href="#5-how-to-run-locally">How to Use</a> |
|
260 |
+
<a href="#6-license">License</a> |
|
261 |
+
</p>
|
262 |
+
|
263 |
+
|
264 |
+
<p align="center">
|
265 |
+
<a href="https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/paper.pdf"><b>Paper Link</b>👁️</a>
|
266 |
+
</p>
|
267 |
+
|
268 |
+
# DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence
|
269 |
+
|
270 |
+
## 1. Introduction
|
271 |
+
We present DeepSeek-Coder-V2, an open-source Mixture-of-Experts (MoE) code language model that achieves performance comparable to GPT4-Turbo in code-specific tasks. Specifically, DeepSeek-Coder-V2 is further pre-trained from DeepSeek-Coder-V2-Base with 6 trillion tokens sourced from a high-quality and multi-source corpus. Through this continued pre-training, DeepSeek-Coder-V2 substantially enhances the coding and mathematical reasoning capabilities of DeepSeek-Coder-V2-Base, while maintaining comparable performance in general language tasks. Compared to DeepSeek-Coder, DeepSeek-Coder-V2 demonstrates significant advancements in various aspects of code-related tasks, as well as reasoning and general capabilities. Additionally, DeepSeek-Coder-V2 expands its support for programming languages from 86 to 338, while extending the context length from 16K to 128K.
|
272 |
+
|
273 |
+
<p align="center">
|
274 |
+
<img width="100%" src="https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/figures/performance.png?raw=true">
|
275 |
+
</p>
|
276 |
+
|
277 |
+
In standard benchmark evaluations, DeepSeek-Coder-V2 achieves superior performance compared to closed-source models such as GPT4-Turbo, Claude 3 Opus, and Gemini 1.5 Pro in coding and math benchmarks. The list of supported programming languages can be found in the paper.
|
278 |
+
|
279 |
+
## 2. Model Downloads
|
280 |
+
|
281 |
+
We release the DeepSeek-Coder-V2 with 16B and 236B parameters based on the [DeepSeekMoE](https://arxiv.org/pdf/2401.06066) framework, which has actived parameters of only 2.4B and 21B , including base and instruct models, to the public.
|
282 |
+
|
283 |
+
<div align="center">
|
284 |
+
|
285 |
+
| **Model** | **#Total Params** | **#Active Params** | **Context Length** | **Download** |
|
286 |
+
| :-----------------------------: | :---------------: | :----------------: | :----------------: | :----------------------------------------------------------: |
|
287 |
+
| DeepSeek-Coder-V2-Lite-Base | 16B | 2.4B | 128k | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Base) |
|
288 |
+
| DeepSeek-Coder-V2-Lite-Instruct | 16B | 2.4B | 128k | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct) |
|
289 |
+
| DeepSeek-Coder-V2-Base | 236B | 21B | 128k | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Base) |
|
290 |
+
| DeepSeek-Coder-V2-Instruct | 236B | 21B | 128k | [🤗 HuggingFace](https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct) |
|
291 |
+
|
292 |
+
</div>
|
293 |
+
|
294 |
+
|
295 |
+
## 3. Chat Website
|
296 |
+
|
297 |
+
You can chat with the DeepSeek-Coder-V2 on DeepSeek's official website: [coder.deepseek.com](https://coder.deepseek.com/sign_in)
|
298 |
+
|
299 |
+
## 4. API Platform
|
300 |
+
We also provide OpenAI-Compatible API at DeepSeek Platform: [platform.deepseek.com](https://platform.deepseek.com/). Sign up for over millions of free tokens. And you can also pay-as-you-go at an unbeatable price.
|
301 |
+
|
302 |
+
<p align="center">
|
303 |
+
<img width="40%" src="https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/figures/model_price.jpg?raw=true">
|
304 |
+
</p>
|
305 |
+
|
306 |
+
|
307 |
+
## 5. How to run locally
|
308 |
+
**Here, we provide some examples of how to use DeepSeek-Coder-V2-Lite model. If you want to utilize DeepSeek-Coder-V2 in BF16 format for inference, 80GB*8 GPUs are required.**
|
309 |
+
|
310 |
+
### Inference with Huggingface's Transformers
|
311 |
+
You can directly employ [Huggingface's Transformers](https://github.com/huggingface/transformers) for model inference.
|
312 |
+
|
313 |
+
#### Code Completion
|
314 |
+
```python
|
315 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
316 |
+
import torch
|
317 |
+
tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-Coder-V2-Lite-Base", trust_remote_code=True)
|
318 |
+
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-Coder-V2-Lite-Base", trust_remote_code=True, torch_dtype=torch.bfloat16).cuda()
|
319 |
+
input_text = "#write a quick sort algorithm"
|
320 |
+
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
|
321 |
+
outputs = model.generate(**inputs, max_length=128)
|
322 |
+
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
323 |
+
```
|
324 |
+
|
325 |
+
#### Code Insertion
|
326 |
+
```python
|
327 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
328 |
+
import torch
|
329 |
+
tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-Coder-V2-Lite-Base", trust_remote_code=True)
|
330 |
+
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-Coder-V2-Lite-Base", trust_remote_code=True, torch_dtype=torch.bfloat16).cuda()
|
331 |
+
input_text = """<|fim▁begin|>def quick_sort(arr):
|
332 |
+
if len(arr) <= 1:
|
333 |
+
return arr
|
334 |
+
pivot = arr[0]
|
335 |
+
left = []
|
336 |
+
right = []
|
337 |
+
<|fim▁hole|>
|
338 |
+
if arr[i] < pivot:
|
339 |
+
left.append(arr[i])
|
340 |
+
else:
|
341 |
+
right.append(arr[i])
|
342 |
+
return quick_sort(left) + [pivot] + quick_sort(right)<|fim▁end|>"""
|
343 |
+
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
|
344 |
+
outputs = model.generate(**inputs, max_length=128)
|
345 |
+
print(tokenizer.decode(outputs[0], skip_special_tokens=True)[len(input_text):])
|
346 |
+
```
|
347 |
+
|
348 |
+
#### Chat Completion
|
349 |
+
|
350 |
+
```python
|
351 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
352 |
+
import torch
|
353 |
+
tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct", trust_remote_code=True)
|
354 |
+
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct", trust_remote_code=True, torch_dtype=torch.bfloat16).cuda()
|
355 |
+
messages=[
|
356 |
+
{ 'role': 'user', 'content': "write a quick sort algorithm in python."}
|
357 |
+
]
|
358 |
+
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
|
359 |
+
# tokenizer.eos_token_id is the id of <|EOT|> token
|
360 |
+
outputs = model.generate(inputs, max_new_tokens=512, do_sample=False, top_k=50, top_p=0.95, num_return_sequences=1, eos_token_id=tokenizer.eos_token_id)
|
361 |
+
print(tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True))
|
362 |
+
```
|
363 |
+
|
364 |
+
|
365 |
+
|
366 |
+
The complete chat template can be found within `tokenizer_config.json` located in the huggingface model repository.
|
367 |
+
|
368 |
+
An example of chat template is as belows:
|
369 |
+
|
370 |
+
```bash
|
371 |
+
<|begin▁of▁sentence|>User: {user_message_1}
|
372 |
+
|
373 |
+
Assistant: {assistant_message_1}<|end▁of▁sentence|>User: {user_message_2}
|
374 |
+
|
375 |
+
Assistant:
|
376 |
+
```
|
377 |
+
|
378 |
+
You can also add an optional system message:
|
379 |
+
|
380 |
+
```bash
|
381 |
+
<|begin▁of▁sentence|>{system_message}
|
382 |
+
|
383 |
+
User: {user_message_1}
|
384 |
+
|
385 |
+
Assistant: {assistant_message_1}<|end▁of▁sentence|>User: {user_message_2}
|
386 |
+
|
387 |
+
Assistant:
|
388 |
+
```
|
389 |
+
|
390 |
+
### Inference with vLLM (recommended)
|
391 |
+
To utilize [vLLM](https://github.com/vllm-project/vllm) for model inference, please merge this Pull Request into your vLLM codebase: https://github.com/vllm-project/vllm/pull/4650.
|
392 |
+
|
393 |
+
```python
|
394 |
+
from transformers import AutoTokenizer
|
395 |
+
from vllm import LLM, SamplingParams
|
396 |
+
|
397 |
+
max_model_len, tp_size = 8192, 1
|
398 |
+
model_name = "deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct"
|
399 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
400 |
+
llm = LLM(model=model_name, tensor_parallel_size=tp_size, max_model_len=max_model_len, trust_remote_code=True, enforce_eager=True)
|
401 |
+
sampling_params = SamplingParams(temperature=0.3, max_tokens=256, stop_token_ids=[tokenizer.eos_token_id])
|
402 |
+
|
403 |
+
messages_list = [
|
404 |
+
[{"role": "user", "content": "Who are you?"}],
|
405 |
+
[{"role": "user", "content": "write a quick sort algorithm in python."}],
|
406 |
+
[{"role": "user", "content": "Write a piece of quicksort code in C++."}],
|
407 |
+
]
|
408 |
+
|
409 |
+
prompt_token_ids = [tokenizer.apply_chat_template(messages, add_generation_prompt=True) for messages in messages_list]
|
410 |
+
|
411 |
+
outputs = llm.generate(prompt_token_ids=prompt_token_ids, sampling_params=sampling_params)
|
412 |
+
|
413 |
+
generated_text = [output.outputs[0].text for output in outputs]
|
414 |
+
print(generated_text)
|
415 |
+
```
|
416 |
+
|
417 |
+
|
418 |
+
|
419 |
+
## 6. License
|
420 |
+
|
421 |
+
This code repository is licensed under [the MIT License](https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/LICENSE-CODE). The use of DeepSeek-Coder-V2 Base/Instruct models is subject to [the Model License](https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/LICENSE-MODEL). DeepSeek-Coder-V2 series (including Base and Instruct) supports commercial use.
|
422 |
+
|
423 |
+
|
424 |
+
## 7. Contact
|
425 |
+
If you have any questions, please raise an issue or contact us at [[email protected]]([email protected]).
|