Skip to content

Commit 0bbbf2a

Browse files
Amitavoogemini-code-assist[bot]sachinprasadhs
authored
Register Qwen2.5-Coder presets (keras-team#2547)
* Add Qwen-2.5-Coder 0.5B preset * Fix formatting for Qwen-2.5-Coder preset * Add Qwen2.5-Coder-1.5B preset to Qwen conversion * Apply suggestion from @gemini-code-assist[bot] Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> * Update Qwen2.5 Coder presets and checkpoint mapping * Fix duplicate qwen2.5_coder_1.5b_en preset entry * Add Qwen 2.5 coder presets and update Kaggle handles * nit --------- Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: Sachin Prasad <sachinprasad@google.com>
1 parent befeaed commit 0bbbf2a

2 files changed

Lines changed: 169 additions & 1 deletion

File tree

‎keras_hub/src/models/qwen/qwen_presets.py‎

Lines changed: 156 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -58,4 +58,160 @@
5858
},
5959
"kaggle_handle": "kaggle://keras/qwen/keras/qwen2.5_instruct_72b_en/2",
6060
},
61+
"qwen2.5_coder_0.5b": {
62+
"metadata": {
63+
"description": (
64+
"Code-focused fine-tuned Qwen-2.5 model with 0.5 "
65+
"billion parameters."
66+
),
67+
"params": 494032768,
68+
"path": "qwen",
69+
},
70+
"kaggle_handle": (
71+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_0.5b/1"
72+
),
73+
},
74+
"qwen2.5_coder_1.5b": {
75+
"metadata": {
76+
"description": (
77+
"Code-focused fine-tuned 28-layer Qwen-2.5 model with 1.5 "
78+
"billion parameters."
79+
),
80+
"params": 1543434240,
81+
"path": "qwen",
82+
},
83+
"kaggle_handle": (
84+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_1.5b/1"
85+
),
86+
},
87+
"qwen2.5_coder_3b": {
88+
"metadata": {
89+
"description": (
90+
"Code-focused fine-tuned Qwen-2.5 model with 3 "
91+
"billion parameters."
92+
),
93+
"params": 3085938688,
94+
"path": "qwen",
95+
},
96+
"kaggle_handle": (
97+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_3b/1"
98+
),
99+
},
100+
"qwen2.5_coder_7b": {
101+
"metadata": {
102+
"description": (
103+
"Code-focused fine-tuned Qwen-2.5 model with 7 "
104+
"billion parameters."
105+
),
106+
"params": 6993420288,
107+
"path": "qwen",
108+
},
109+
"kaggle_handle": (
110+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_7b/1"
111+
),
112+
},
113+
"qwen2.5_coder_14b": {
114+
"metadata": {
115+
"description": (
116+
"Code-focused fine-tuned Qwen-2.5 model with 14 "
117+
"billion parameters."
118+
),
119+
"params": 14000000000,
120+
"path": "qwen",
121+
},
122+
"kaggle_handle": (
123+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_14b/1"
124+
),
125+
},
126+
"qwen2.5_coder_32b": {
127+
"metadata": {
128+
"description": (
129+
"Code-focused fine-tuned Qwen-2.5 model with 32 "
130+
"billion parameters."
131+
),
132+
"params": 32763876352,
133+
"path": "qwen",
134+
},
135+
"kaggle_handle": (
136+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_32b/1"
137+
),
138+
},
139+
"qwen2.5_coder_instruct_0.5b": {
140+
"metadata": {
141+
"description": (
142+
"Instruction-tuned code-focused Qwen-2.5 model with "
143+
"0.5 billion parameters."
144+
),
145+
"params": 494032768,
146+
"path": "qwen",
147+
},
148+
"kaggle_handle": (
149+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_instruct_0.5b/1"
150+
),
151+
},
152+
"qwen2.5_coder_instruct_1.5b": {
153+
"metadata": {
154+
"description": (
155+
"Instruction-tuned code-focused Qwen-2.5 model with "
156+
"1.5 billion parameters."
157+
),
158+
"params": 1543434240,
159+
"path": "qwen",
160+
},
161+
"kaggle_handle": (
162+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_instruct_1.5b/1"
163+
),
164+
},
165+
"qwen2.5_coder_instruct_3b": {
166+
"metadata": {
167+
"description": (
168+
"Instruction-tuned code-focused Qwen-2.5 model with "
169+
"3 billion parameters."
170+
),
171+
"params": 3085938688,
172+
"path": "qwen",
173+
},
174+
"kaggle_handle": (
175+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_instruct_3b/1"
176+
),
177+
},
178+
"qwen2.5_coder_instruct_7b": {
179+
"metadata": {
180+
"description": (
181+
"Instruction-tuned code-focused Qwen-2.5 model with "
182+
"7 billion parameters."
183+
),
184+
"params": 6993420288,
185+
"path": "qwen",
186+
},
187+
"kaggle_handle": (
188+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_instruct_7b/1"
189+
),
190+
},
191+
"qwen2.5_coder_instruct_14b": {
192+
"metadata": {
193+
"description": (
194+
"Instruction-tuned code-focused Qwen-2.5 model with "
195+
"14 billion parameters."
196+
),
197+
"params": 14000000000,
198+
"path": "qwen",
199+
},
200+
"kaggle_handle": (
201+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_instruct_14b/1"
202+
),
203+
},
204+
"qwen2.5_coder_instruct_32b": {
205+
"metadata": {
206+
"description": (
207+
"Instruction-tuned code-focused Qwen-2.5 model with "
208+
"32 billion parameters."
209+
),
210+
"params": 32763876352,
211+
"path": "qwen",
212+
},
213+
"kaggle_handle": (
214+
"kaggle://keras/qwen2-5-coder/keras/qwen2.5_coder_instruct_32b/1"
215+
),
216+
},
61217
}

‎tools/checkpoint_conversion/convert_qwen_checkpoints.py‎

Lines changed: 13 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -21,11 +21,23 @@
2121

2222
PRESET_MAP = {
2323
"qwen2.5_0.5b_en": "Qwen/Qwen2.5-0.5B",
24-
"qwen2.5_7b_en": "Qwen/Qwen2.5-7B",
2524
"qwen2.5_3b_en": "Qwen/Qwen2.5-3B",
25+
"qwen2.5_7b_en": "Qwen/Qwen2.5-7B",
2626
"qwen2.5_instruct_0.5b_en": "Qwen/Qwen2.5-0.5B-Instruct",
2727
"qwen2.5_instruct_32b_en": "Qwen/Qwen2.5-32B-Instruct",
2828
"qwen2.5_instruct_72b_en": "Qwen/Qwen2.5-72B-Instruct",
29+
"qwen2.5_coder_0.5b": "Qwen/Qwen2.5-Coder-0.5B",
30+
"qwen2.5_coder_1.5b": "Qwen/Qwen2.5-Coder-1.5B",
31+
"qwen2.5_coder_3b": "Qwen/Qwen2.5-Coder-3B",
32+
"qwen2.5_coder_7b": "Qwen/Qwen2.5-Coder-7B",
33+
"qwen2.5_coder_14b": "Qwen/Qwen2.5-Coder-14B",
34+
"qwen2.5_coder_32b": "Qwen/Qwen2.5-Coder-32B",
35+
"qwen2.5_coder_instruct_0.5b": "Qwen/Qwen2.5-Coder-0.5B-Instruct",
36+
"qwen2.5_coder_instruct_1.5b": "Qwen/Qwen2.5-Coder-1.5B-Instruct",
37+
"qwen2.5_coder_instruct_3b": "Qwen/Qwen2.5-Coder-3B-Instruct",
38+
"qwen2.5_coder_instruct_7b": "Qwen/Qwen2.5-Coder-7B-Instruct",
39+
"qwen2.5_coder_instruct_14b": "Qwen/Qwen2.5-Coder-14B-Instruct",
40+
"qwen2.5_coder_instruct_32b": "Qwen/Qwen2.5-Coder-32B-Instruct",
2941
}
3042

3143
FLAGS = flags.FLAGS

0 commit comments

Comments
 (0)