Skip to content

Commit 34037f3

Browse files
cebtenzzrecebtenzzre
authored
models: add DeepSeek-R1 distillations to official models list (#3437)
Signed-off-by: Jared Van Bortel <[email protected]>
1 parent 007a7af commit 34037f3

File tree

3 files changed

+73
-3
lines changed

3 files changed

+73
-3
lines changed

‎gpt4all-chat/CHANGELOG.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -9,6 +9,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
99
### Added
1010
- Support DeepSeek-R1 Qwen models ([#3431](https://github.com/nomic-ai/gpt4all/pull/3431))
1111
- Support for think tags in the GUI ([#3440](https://github.com/nomic-ai/gpt4all/pull/3440))
12+
- Support specifying SHA256 hash in models3.json instead of MD5 ([#3437](https://github.com/nomic-ai/gpt4all/pull/3437))
1213

1314
### Changed
1415
- Use minja instead of Jinja2Cpp for significantly improved template compatibility ([#3433](https://github.com/nomic-ai/gpt4all/pull/3433))

‎gpt4all-chat/metadata/models3.json‎

Lines changed: 61 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -32,6 +32,66 @@
3232
"systemPrompt": "",
3333
"chatTemplate": "{%- set loop_messages = messages %}\n{%- for message in loop_messages %}\n {%- set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>\\n\\n'+ message['content'] | trim + '<|eot_id|>' %}\n {{- content }}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|start_header_id|>assistant<|end_header_id|>\\n\\n' }}\n{%- endif %}"
3434
},
35+
{
36+
"order": "aa1",
37+
"sha256sum": "5cd4ee65211770f1d99b4f6f4951780b9ef40e29314bd6542bb5bd0ad0bc29d1",
38+
"name": "DeepSeek-R1-Distill-Qwen-7B",
39+
"filename": "DeepSeek-R1-Distill-Qwen-7B-Q4_0.gguf",
40+
"filesize": "4444121056",
41+
"requires": "3.8.0",
42+
"ramrequired": "8",
43+
"parameters": "7 billion",
44+
"quant": "q4_0",
45+
"type": "deepseek",
46+
"description": "<p>The official Qwen2.5-Math-7B distillation of DeepSeek-R1.</p><ul><li>License: <a href=\"https://opensource.org/license/mit\">MIT</a></li><li>No restrictions on commercial use</li><li>#reasoning</li></ul>",
47+
"url": "https://huggingface.co/bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF/resolve/main/DeepSeek-R1-Distill-Qwen-7B-Q4_0.gguf",
48+
"chatTemplate": "{%- if not add_generation_prompt is defined %}\n {%- set add_generation_prompt = false %}\n{%- endif %}\n{%- if messages[0]['role'] == 'system' %}\n {{- messages[0]['content'] }}\n{%- endif %}\n{%- for message in messages %}\n {%- if message['role'] == 'user' %}\n {{- '<｜User｜>' + message['content'] }}\n {%- endif %}\n {%- if message['role'] == 'assistant' %}\n {%- set content = message['content'] | regex_replace('^[\\\\s\\\\S]*</think>', '') %}\n {{- '<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>' }}\n {%- endif %}\n{%- endfor -%}\n{%- if add_generation_prompt %}\n {{- '<｜Assistant｜>' }}\n{%- endif %}"
49+
},
50+
{
51+
"order": "aa2",
52+
"sha256sum": "906b3382f2680f4ce845459b4a122e904002b075238080307586bcffcde49eef",
53+
"name": "DeepSeek-R1-Distill-Qwen-14B",
54+
"filename": "DeepSeek-R1-Distill-Qwen-14B-Q4_0.gguf",
55+
"filesize": "8544267680",
56+
"requires": "3.8.0",
57+
"ramrequired": "16",
58+
"parameters": "14 billion",
59+
"quant": "q4_0",
60+
"type": "deepseek",
61+
"description": "<p>The official Qwen2.5-14B distillation of DeepSeek-R1.</p><ul><li>License: <a href=\"https://opensource.org/license/mit\">MIT</a></li><li>No restrictions on commercial use</li><li>#reasoning</li></ul>",
62+
"url": "https://huggingface.co/bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF/resolve/main/DeepSeek-R1-Distill-Qwen-14B-Q4_0.gguf",
63+
"chatTemplate": "{%- if not add_generation_prompt is defined %}\n {%- set add_generation_prompt = false %}\n{%- endif %}\n{%- if messages[0]['role'] == 'system' %}\n {{- messages[0]['content'] }}\n{%- endif %}\n{%- for message in messages %}\n {%- if message['role'] == 'user' %}\n {{- '<｜User｜>' + message['content'] }}\n {%- endif %}\n {%- if message['role'] == 'assistant' %}\n {%- set content = message['content'] | regex_replace('^[\\\\s\\\\S]*</think>', '') %}\n {{- '<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>' }}\n {%- endif %}\n{%- endfor -%}\n{%- if add_generation_prompt %}\n {{- '<｜Assistant｜>' }}\n{%- endif %}"
64+
},
65+
{
66+
"order": "aa3",
67+
"sha256sum": "0eb93e436ac8beec18aceb958c120d282cb2cf5451b23185e7be268fe9d375cc",
68+
"name": "DeepSeek-R1-Distill-Llama-8B",
69+
"filename": "DeepSeek-R1-Distill-Llama-8B-Q4_0.gguf",
70+
"filesize": "4675894112",
71+
"requires": "3.8.0",
72+
"ramrequired": "8",
73+
"parameters": "8 billion",
74+
"quant": "q4_0",
75+
"type": "deepseek",
76+
"description": "<p>The official Llama-3.1-8B distillation of DeepSeek-R1.</p><ul><li>License: <a href=\"https://opensource.org/license/mit\">MIT</a></li><li>No restrictions on commercial use</li><li>#reasoning</li></ul>",
77+
"url": "https://huggingface.co/bartowski/DeepSeek-R1-Distill-Llama-8B-GGUF/resolve/main/DeepSeek-R1-Distill-Llama-8B-Q4_0.gguf",
78+
"chatTemplate": "{%- if not add_generation_prompt is defined %}\n {%- set add_generation_prompt = false %}\n{%- endif %}\n{%- if messages[0]['role'] == 'system' %}\n {{- messages[0]['content'] }}\n{%- endif %}\n{%- for message in messages %}\n {%- if message['role'] == 'user' %}\n {{- '<｜User｜>' + message['content'] }}\n {%- endif %}\n {%- if message['role'] == 'assistant' %}\n {%- set content = message['content'] | regex_replace('^[\\\\s\\\\S]*</think>', '') %}\n {{- '<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>' }}\n {%- endif %}\n{%- endfor -%}\n{%- if add_generation_prompt %}\n {{- '<｜Assistant｜>' }}\n{%- endif %}"
79+
},
80+
{
81+
"order": "aa4",
82+
"sha256sum": "b3af887d0a015b39fab2395e4faf682c1a81a6a3fd09a43f0d4292f7d94bf4d0",
83+
"name": "DeepSeek-R1-Distill-Qwen-1.5B",
84+
"filename": "DeepSeek-R1-Distill-Qwen-1.5B-Q4_0.gguf",
85+
"filesize": "1068807776",
86+
"requires": "3.8.0",
87+
"ramrequired": "3",
88+
"parameters": "1.5 billion",
89+
"quant": "q4_0",
90+
"type": "deepseek",
91+
"description": "<p>The official Qwen2.5-Math-1.5B distillation of DeepSeek-R1.</p><ul><li>License: <a href=\"https://opensource.org/license/mit\">MIT</a></li><li>No restrictions on commercial use</li><li>#reasoning</li></ul>",
92+
"url": "https://huggingface.co/bartowski/DeepSeek-R1-Distill-Qwen-1.5B-GGUF/resolve/main/DeepSeek-R1-Distill-Qwen-1.5B-Q4_0.gguf",
93+
"chatTemplate": "{%- if not add_generation_prompt is defined %}\n {%- set add_generation_prompt = false %}\n{%- endif %}\n{%- if messages[0]['role'] == 'system' %}\n {{- messages[0]['content'] }}\n{%- endif %}\n{%- for message in messages %}\n {%- if message['role'] == 'user' %}\n {{- '<｜User｜>' + message['content'] }}\n {%- endif %}\n {%- if message['role'] == 'assistant' %}\n {%- set content = message['content'] | regex_replace('^[\\\\s\\\\S]*</think>', '') %}\n {{- '<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>' }}\n {%- endif %}\n{%- endfor -%}\n{%- if add_generation_prompt %}\n {{- '<｜Assistant｜>' }}\n{%- endif %}"
94+
},
3595
{
3696
"order": "b",
3797
"md5sum": "27b44e8ae1817525164ddf4f8dae8af4",
@@ -472,7 +532,7 @@
472532
"filename": "qwen2-1_5b-instruct-q4_0.gguf",
473533
"filesize": "937532800",
474534
"requires": "3.0",
475-
"ramrequired": "4",
535+
"ramrequired": "3",
476536
"parameters": "1.5 billion",
477537
"quant": "q4_0",
478538
"type": "qwen2",

‎gpt4all-chat/src/modellist.cpp‎

Lines changed: 11 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1621,7 +1621,6 @@ void ModelList::parseModelsJsonFile(const QByteArray &jsonData, bool save)
16211621
QString requiresVersion = obj["requires"].toString();
16221622
QString versionRemoved = obj["removedIn"].toString();
16231623
QString url = obj["url"].toString();
1624-
QByteArray modelHash = obj["md5sum"].toString().toLatin1();
16251624
bool isDefault = obj.contains("isDefault") && obj["isDefault"] == u"true"_s;
16261625
bool disableGUI = obj.contains("disableGUI") && obj["disableGUI"] == u"true"_s;
16271626
QString description = obj["description"].toString();
@@ -1632,6 +1631,16 @@ void ModelList::parseModelsJsonFile(const QByteArray &jsonData, bool save)
16321631
QString type = obj["type"].toString();
16331632
bool isEmbeddingModel = obj["embeddingModel"].toBool();
16341633

1634+
QByteArray modelHash;
1635+
ModelInfo::HashAlgorithm hashAlgorithm;
1636+
if (auto it = obj.find("sha256sum"_L1); it != obj.end()) {
1637+
modelHash = it->toString().toLatin1();
1638+
hashAlgorithm = ModelInfo::Sha256;
1639+
} else {
1640+
modelHash = obj["md5sum"].toString().toLatin1();
1641+
hashAlgorithm = ModelInfo::Md5;
1642+
}
1643+
16351644
// Some models aren't supported in the GUI at all
16361645
if (disableGUI)
16371646
continue;
@@ -1660,7 +1669,7 @@ void ModelList::parseModelsJsonFile(const QByteArray &jsonData, bool save)
16601669
{ ModelList::FilenameRole, modelFilename },
16611670
{ ModelList::FilesizeRole, modelFilesize },
16621671
{ ModelList::HashRole, modelHash },
1663-
{ ModelList::HashAlgorithmRole, ModelInfo::Md5 },
1672+
{ ModelList::HashAlgorithmRole, hashAlgorithm },
16641673
{ ModelList::DefaultRole, isDefault },
16651674
{ ModelList::DescriptionRole, description },
16661675
{ ModelList::RequiresVersionRole, requiresVersion },

0 commit comments

Comments
 (0)