Skip to content

Commit 2fb0e29

Browse files
giladgdgiladgd
committed
chore: update recommended models
1 parent 06d23e7 commit 2fb0e29

File tree

1 file changed

+77
-19
lines changed

1 file changed

+77
-19
lines changed

‎src/cli/recommendedModels.ts‎

Lines changed: 77 additions & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -1,21 +1,79 @@
11
import {ModelRecommendation} from "./utils/resolveModelRecommendationFileOptions.js";
22

33
export const recommendedModels: ModelRecommendation[] = [{
4+
name: "DeepSeek R1 Distill Qwen 7B",
5+
abilities: ["chat", "complete", "functionCalling", "reasoning"],
6+
description: "DeepSeek R1 model was created by DeepSeek and is using chain of though (CoT) to reason across a wide variety of topics.\n" +
7+
"It's optimized for an assistant-like chat use cases, with support for function calling.\n" +
8+
"This model is censored, but its responses quality on many topics is extremely high.\n" +
9+
"This is the 7 billion parameters version of the model - a fine tuned Qwen 2.5 7B base model with distillation from the 671B DeepSeek R1 version.",
10+
11+
fileOptions: [
12+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q8_0",
13+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q6_K",
14+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q5_K_M",
15+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q5_K_S",
16+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M"
17+
]
18+
}, {
419
name: "DeepSeek R1 Distill Qwen 14B",
520
abilities: ["chat", "complete", "functionCalling", "reasoning"],
621
description: "DeepSeek R1 model was created by DeepSeek and is using chain of though (CoT) to reason across a wide variety of topics.\n" +
722
"It's optimized for an assistant-like chat use cases, with support for function calling.\n" +
823
"This model is censored, but its responses quality on many topics is extremely high.\n" +
9-
"This is the 14 billion parameters version of the model - a fine tuned Qwen 2.5 14B base model with distillation from the 617B DeepSeek R1 version.",
24+
"This is the 14 billion parameters version of the model - a fine tuned Qwen 2.5 14B base model with distillation from the 671B DeepSeek R1 version.",
25+
26+
fileOptions: [
27+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q8_0",
28+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q6_K",
29+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q5_K_M",
30+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q5_K_S",
31+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q4_K_M"
32+
]
33+
}, {
34+
name: "DeepSeek R1 Distill Qwen 32B",
35+
abilities: ["chat", "complete", "functionCalling", "reasoning"],
36+
description: "DeepSeek R1 model was created by DeepSeek and is using chain of though (CoT) to reason across a wide variety of topics.\n" +
37+
"It's optimized for an assistant-like chat use cases, with support for function calling.\n" +
38+
"This model is censored, but its responses quality on many topics is extremely high.\n" +
39+
"This is the 32 billion parameters version of the model - a fine tuned Qwen 2.5 32B base model with distillation from the 671B DeepSeek R1 version.",
40+
41+
fileOptions: [
42+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q8_0",
43+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q6_K",
44+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q5_K_M",
45+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q5_K_S",
46+
"hf:mradermacher/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q4_K_M"
47+
]
48+
}, {
49+
name: "DeepSeek R1 Distill Llama 8B",
50+
abilities: ["chat", "complete", "functionCalling", "reasoning"],
51+
description: "DeepSeek R1 model was created by DeepSeek and is using chain of though (CoT) to reason across a wide variety of topics.\n" +
52+
"It's optimized for an assistant-like chat use cases, with support for function calling.\n" +
53+
"This model is censored, even though it's based on Llama 3.1.\n" +
54+
"This is the 8 billion parameters version of the model - a fine tuned Llama 3.1 8B base model with distillation from the 671B DeepSeek R1 version.",
55+
56+
fileOptions: [
57+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-8B-GGUF:Q8_0",
58+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-8B-GGUF:Q6_K",
59+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-8B-GGUF:Q5_K_M",
60+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-8B-GGUF:Q5_K_S",
61+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-8B-GGUF:Q4_K_M"
62+
]
63+
}, {
64+
name: "DeepSeek R1 Distill Llama 70B",
65+
abilities: ["chat", "complete", "functionCalling", "reasoning"],
66+
description: "DeepSeek R1 model was created by DeepSeek and is using chain of though (CoT) to reason across a wide variety of topics.\n" +
67+
"It's optimized for an assistant-like chat use cases, with support for function calling.\n" +
68+
"This model is censored, even though it's based on Llama 3.3.\n" +
69+
"This is the 70 billion parameters version of the model - a fine tuned Llama 3.3 70B base model with distillation from the 671B DeepSeek R1 version.",
1070

1171
fileOptions: [
12-
"hf:bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q8_0",
13-
"hf:bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q6_K_L",
14-
"hf:bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q6_K",
15-
"hf:bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q5_K_L",
16-
"hf:bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q5_K_M",
17-
"hf:bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q5_K_S",
18-
"hf:bartowski/DeepSeek-R1-Distill-Qwen-14B-GGUF:Q4_K_M"
72+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-70B-GGUF/DeepSeek-R1-Distill-Llama-70B.Q8_0.gguf.part1of2",
73+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-70B-GGUF/DeepSeek-R1-Distill-Llama-70B.Q6_K.gguf.part1of2",
74+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-70B-GGUF:Q5_K_M",
75+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-70B-GGUF:Q5_K_S",
76+
"hf:mradermacher/DeepSeek-R1-Distill-Llama-70B-GGUF:Q4_K_M"
1977
]
2078
}, {
2179
name: "Llama 3.1 8B",
@@ -28,17 +86,6 @@ export const recommendedModels: ModelRecommendation[] = [{
2886
"hf:mradermacher/Meta-Llama-3.1-8B-Instruct-GGUF:Q6_K",
2987
"hf:mradermacher/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M"
3088
]
31-
}, {
32-
name: "Phi 4 14B",
33-
abilities: ["chat", "complete", "functionCalling"],
34-
description: "Phi 4 model was created by Microsoft and is optimized for complex reasoning in areas such as math.",
35-
36-
fileOptions: [
37-
"hf:mradermacher/phi-4-GGUF:Q8_0",
38-
"hf:mradermacher/phi-4-GGUF:Q6_K",
39-
"hf:mradermacher/phi-4-GGUF:Q4_K_M",
40-
"hf:mradermacher/phi-4-GGUF:Q4_K_S"
41-
]
4289
}, {
4390
name: "Llama 3.1 70B",
4491
abilities: ["chat", "complete", "functionCalling"],
@@ -63,6 +110,17 @@ export const recommendedModels: ModelRecommendation[] = [{
63110
"hf:mradermacher/Meta-Llama-3.1-405B-Instruct-GGUF/Meta-Llama-3.1-405B-Instruct.Q3_K_L.gguf.part1of5",
64111
"hf:mradermacher/Meta-Llama-3.1-405B-Instruct-GGUF/Meta-Llama-3.1-405B-Instruct.Q3_K_M.gguf.part1of4"
65112
]
113+
}, {
114+
name: "Phi 4 14B",
115+
abilities: ["chat", "complete", "functionCalling"],
116+
description: "Phi 4 model was created by Microsoft and is optimized for complex reasoning in areas such as math.",
117+
118+
fileOptions: [
119+
"hf:mradermacher/phi-4-GGUF:Q8_0",
120+
"hf:mradermacher/phi-4-GGUF:Q6_K",
121+
"hf:mradermacher/phi-4-GGUF:Q4_K_M",
122+
"hf:mradermacher/phi-4-GGUF:Q4_K_S"
123+
]
66124
}, {
67125
name: "Mistral Nemo 12B",
68126
abilities: ["chat", "complete", "functionCalling"],

0 commit comments

Comments
 (0)