lemonade-sdk 8.1.12__py3-none-any.whl → 8.2.0__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Potentially problematic release.

This version of lemonade-sdk might be problematic. Click here for more details.

Files changed (24) hide show
  1. lemonade/common/system_info.py +0 -26
  2. lemonade/tools/flm/utils.py +70 -22
  3. lemonade/tools/llamacpp/load.py +10 -1
  4. lemonade/tools/llamacpp/utils.py +82 -8
  5. lemonade/tools/oga/load.py +38 -142
  6. lemonade/tools/oga/migration.py +403 -0
  7. lemonade/tools/server/llamacpp.py +20 -1
  8. lemonade/tools/server/serve.py +334 -16
  9. lemonade/tools/server/static/js/models.js +416 -18
  10. lemonade/tools/server/static/js/shared.js +41 -4
  11. lemonade/tools/server/static/styles.css +204 -0
  12. lemonade/tools/server/static/webapp.html +32 -0
  13. lemonade/version.py +1 -1
  14. lemonade_install/install.py +33 -579
  15. {lemonade_sdk-8.1.12.dist-info → lemonade_sdk-8.2.0.dist-info}/METADATA +4 -3
  16. {lemonade_sdk-8.1.12.dist-info → lemonade_sdk-8.2.0.dist-info}/RECORD +24 -23
  17. lemonade_server/cli.py +10 -0
  18. lemonade_server/model_manager.py +172 -11
  19. lemonade_server/server_models.json +94 -71
  20. {lemonade_sdk-8.1.12.dist-info → lemonade_sdk-8.2.0.dist-info}/WHEEL +0 -0
  21. {lemonade_sdk-8.1.12.dist-info → lemonade_sdk-8.2.0.dist-info}/entry_points.txt +0 -0
  22. {lemonade_sdk-8.1.12.dist-info → lemonade_sdk-8.2.0.dist-info}/licenses/LICENSE +0 -0
  23. {lemonade_sdk-8.1.12.dist-info → lemonade_sdk-8.2.0.dist-info}/licenses/NOTICE.md +0 -0
  24. {lemonade_sdk-8.1.12.dist-info → lemonade_sdk-8.2.0.dist-info}/top_level.txt +0 -0
lemonade_server/server_models.json CHANGED
@@ -44,151 +44,160 @@
44
44
  "size": 5.78
45
45
  },
46
46
  "Llama-3.2-1B-Instruct-Hybrid": {
47
- "checkpoint": "amd/Llama-3.2-1B-Instruct-awq-g128-int4-asym-fp16-onnx-hybrid",
47
+ "checkpoint": "amd/Llama-3.2-1B-Instruct-onnx-ryzenai-hybrid",
48
48
  "recipe": "oga-hybrid",
49
49
  "suggested": true,
50
- "size": 1.75
50
+ "size": 1.89
51
51
  },
52
52
  "Llama-3.2-3B-Instruct-Hybrid": {
53
- "checkpoint": "amd/Llama-3.2-3B-Instruct-awq-g128-int4-asym-fp16-onnx-hybrid",
53
+ "checkpoint": "amd/Llama-3.2-3B-Instruct-onnx-ryzenai-hybrid",
54
54
  "recipe": "oga-hybrid",
55
55
  "suggested": true,
56
- "size": 3.97
56
+ "size": 4.28
57
57
  },
58
58
  "Phi-3-Mini-Instruct-Hybrid": {
59
- "checkpoint": "amd/Phi-3-mini-4k-instruct-awq-g128-int4-asym-fp16-onnx-hybrid",
59
+ "checkpoint": "amd/Phi-3-mini-4k-instruct-onnx-ryzenai-hybrid",
60
60
  "recipe": "oga-hybrid",
61
61
  "suggested": true,
62
- "size": 3.89
62
+ "size": 4.18
63
63
  },
64
64
  "Phi-3.5-Mini-Instruct-Hybrid": {
65
- "checkpoint": "amd/Phi-3.5-mini-instruct-awq-g128-int4-asym-fp16-onnx-hybrid",
65
+ "checkpoint": "amd/Phi-3.5-mini-instruct-onnx-ryzenai-hybrid",
66
66
  "recipe": "oga-hybrid",
67
67
  "suggested": false,
68
- "size": 3.92
68
+ "size": 4.21
69
69
  },
70
70
  "Qwen-1.5-7B-Chat-Hybrid": {
71
- "checkpoint": "amd/Qwen1.5-7B-Chat-awq-g128-int4-asym-fp16-onnx-hybrid",
71
+ "checkpoint": "amd/Qwen1.5-7B-Chat-onnx-ryzenai-hybrid",
72
72
  "recipe": "oga-hybrid",
73
73
  "suggested": true,
74
- "size": 8.22
74
+ "size": 8.83
75
75
  },
76
76
  "Qwen-2.5-7B-Instruct-Hybrid": {
77
- "checkpoint": "amd/Qwen2.5-7B-Instruct-awq-uint4-asym-g128-lmhead-g32-fp16-onnx-hybrid",
77
+ "checkpoint": "amd/Qwen2.5-7B-Instruct-onnx-ryzenai-hybrid",
78
78
  "recipe": "oga-hybrid",
79
79
  "suggested": true,
80
- "size": 8.42
80
+ "size": 8.65
81
81
  },
82
82
  "Qwen-2.5-3B-Instruct-Hybrid": {
83
- "checkpoint": "amd/Qwen2.5-3B-Instruct-awq-uint4-asym-g128-lmhead-g32-fp16-onnx-hybrid",
83
+ "checkpoint": "amd/Qwen2.5-3B-Instruct-onnx-ryzenai-hybrid",
84
84
  "recipe": "oga-hybrid",
85
85
  "suggested": true,
86
- "size": 3.84
86
+ "size": 3.97
87
87
  },
88
88
  "Qwen-2.5-1.5B-Instruct-Hybrid": {
89
- "checkpoint": "amd/Qwen2.5-1.5B-Instruct-awq-uint4-asym-g128-lmhead-g32-fp16-onnx-hybrid",
89
+ "checkpoint": "amd/Qwen2.5-1.5B-Instruct-onnx-ryzenai-hybrid",
90
90
  "recipe": "oga-hybrid",
91
91
  "suggested": true,
92
- "size": 2.08
92
+ "size": 2.16
93
93
  },
94
94
  "DeepSeek-R1-Distill-Llama-8B-Hybrid": {
95
- "checkpoint": "amd/DeepSeek-R1-Distill-Llama-8B-awq-asym-uint4-g128-lmhead-onnx-hybrid",
95
+ "checkpoint": "amd/DeepSeek-R1-Distill-Llama-8B-onnx-ryzenai-hybrid",
96
96
  "recipe": "oga-hybrid",
97
97
  "suggested": true,
98
98
  "labels": ["reasoning"],
99
- "size": 8.45
99
+ "size": 9.09
100
100
  },
101
101
  "DeepSeek-R1-Distill-Qwen-7B-Hybrid": {
102
- "checkpoint": "amd/DeepSeek-R1-Distill-Qwen-7B-awq-asym-uint4-g128-lmhead-onnx-hybrid",
102
+ "checkpoint": "amd/DeepSeek-R1-Distill-Qwen-7B-onnx-ryzenai-hybrid",
103
103
  "recipe": "oga-hybrid",
104
104
  "max_prompt_length": 2000,
105
105
  "suggested": false,
106
106
  "labels": ["reasoning"],
107
- "size": 8.84
107
+ "size": 8.67
108
108
  },
109
109
  "Mistral-7B-v0.3-Instruct-Hybrid": {
110
- "checkpoint": "amd/Mistral-7B-Instruct-v0.3-awq-g128-int4-asym-fp16-onnx-hybrid",
110
+ "checkpoint": "amd/Mistral-7B-Instruct-v0.3-onnx-ryzenai-hybrid",
111
111
  "recipe": "oga-hybrid",
112
112
  "suggested": true,
113
- "size": 7.31
113
+ "size": 7.85
114
114
  },
115
115
  "Llama-3.1-8B-Instruct-Hybrid": {
116
- "checkpoint": "amd/Llama-3.1-8B-Instruct-awq-asym-uint4-g128-lmhead-onnx-hybrid",
116
+ "checkpoint": "amd/Meta-Llama-3.1-8B-Instruct-onnx-ryzenai-hybrid",
117
+ "recipe": "oga-hybrid",
118
+ "suggested": true,
119
+ "size": 9.09
120
+ },
121
+ "Qwen3-1.7B-Hybrid": {
122
+ "checkpoint": "amd/Qwen3-1.7B-awq-quant-onnx-hybrid",
123
+ "recipe": "oga-hybrid",
124
+ "suggested": true,
125
+ "labels": ["reasoning"],
126
+ "size": 2.55
127
+ },
128
+ "Phi-4-Mini-Instruct-Hybrid": {
129
+ "checkpoint": "amd/Phi-4-mini-instruct-onnx-ryzenai-hybrid",
117
130
  "recipe": "oga-hybrid",
118
131
  "suggested": true,
119
- "size": 8.47
132
+ "size": 5.46
120
133
  },
121
- "Llama-xLAM-2-8b-fc-r-Hybrid": {
122
- "checkpoint": "amd/Llama-xLAM-2-8b-fc-r-awq-g128-int4-asym-bfp16-onnx-hybrid",
134
+ "Qwen3-4B-Hybrid": {
135
+ "checkpoint": "amd/Qwen3-4B-awq-quant-onnx-hybrid",
123
136
  "recipe": "oga-hybrid",
124
137
  "suggested": true,
125
- "size": 8.47
138
+ "labels": ["reasoning"],
139
+ "size": 5.17
140
+ },
141
+ "Qwen3-8B-Hybrid": {
142
+ "checkpoint": "amd/Qwen3-8B-awq-quant-onnx-hybrid",
143
+ "recipe": "oga-hybrid",
144
+ "suggested": true,
145
+ "labels": ["reasoning"],
146
+ "size": 9.42
126
147
  },
127
148
  "Qwen-2.5-7B-Instruct-NPU": {
128
- "checkpoint": "amd/Qwen2.5-7B-Instruct-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
149
+ "checkpoint": "amd/Qwen2.5-7B-Instruct-onnx-ryzenai-npu",
129
150
  "recipe": "oga-npu",
130
151
  "suggested": true,
131
- "size": 10.14
152
+ "size": 8.82
132
153
  },
133
- "Qwen-2.5-1.5B-Instruct-NPU": {
134
- "checkpoint": "amd/Qwen2.5-1.5B-Instruct-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
154
+ "Qwen-2.5-3B-Instruct-NPU": {
155
+ "checkpoint": "amd/Qwen2.5-3B-Instruct-onnx-ryzenai-npu",
135
156
  "recipe": "oga-npu",
136
157
  "suggested": true,
137
- "size": 2.89
158
+ "size": 4.09
138
159
  },
139
160
  "DeepSeek-R1-Distill-Llama-8B-NPU": {
140
- "checkpoint": "amd/DeepSeek-R1-Distill-Llama-8B-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
161
+ "checkpoint": "amd/DeepSeek-R1-Distill-Llama-8B-onnx-ryzenai-npu",
141
162
  "recipe": "oga-npu",
142
163
  "suggested": true,
143
- "size": 10.63
164
+ "size": 9.30
144
165
  },
145
166
  "DeepSeek-R1-Distill-Qwen-7B-NPU": {
146
- "checkpoint": "amd/DeepSeek-R1-Distill-Qwen-7B-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
167
+ "checkpoint": "amd/DeepSeek-R1-Distill-Qwen-7B-onnx-ryzenai-npu",
147
168
  "recipe": "oga-npu",
148
169
  "suggested": false,
149
- "size": 10.3
170
+ "size": 8.87
150
171
  },
151
172
  "DeepSeek-R1-Distill-Qwen-1.5B-NPU": {
152
- "checkpoint": "amd/DeepSeek-R1-Distill-Qwen-1.5B-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
153
- "recipe": "oga-npu",
154
- "suggested": false,
155
- "size": 3.02
156
- },
157
- "Llama-3.2-3B-Instruct-NPU": {
158
- "checkpoint": "amd/Llama-3.2-3B-Instruct-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
173
+ "checkpoint": "amd/DeepSeek-R1-Distill-Qwen-1.5B-onnx-ryzenai-npu",
159
174
  "recipe": "oga-npu",
160
175
  "suggested": false,
161
- "size": 2.46
176
+ "size": 2.30
162
177
  },
163
178
  "Llama-3.2-1B-Instruct-NPU": {
164
- "checkpoint": "amd/Llama-3.2-1B-Instruct-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
179
+ "checkpoint": "amd/Llama-3.2-1B-Instruct-onnx-ryzenai-npu",
165
180
  "recipe": "oga-npu",
166
181
  "suggested": false,
167
- "size": 1.18
182
+ "size": 1.96
168
183
  },
169
184
  "Mistral-7B-v0.3-Instruct-NPU": {
170
- "checkpoint": "amd/Mistral-7B-Instruct-v0.3-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
185
+ "checkpoint": "amd/Mistral-7B-Instruct-v0.3-onnx-ryzenai-npu",
171
186
  "recipe": "oga-npu",
172
187
  "suggested": true,
173
- "size": 11.75
188
+ "size": 8.09
174
189
  },
175
190
  "Phi-3.5-Mini-Instruct-NPU": {
176
- "checkpoint": "amd/Phi-3.5-mini-instruct-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
191
+ "checkpoint": "amd/Phi-3.5-mini-instruct-onnx-ryzenai-npu",
177
192
  "recipe": "oga-npu",
178
193
  "suggested": true,
179
- "size": 4.18
194
+ "size": 4.35
180
195
  },
181
196
  "ChatGLM-3-6b-Instruct-NPU": {
182
- "checkpoint": "amd/chatglm3-6b-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
197
+ "checkpoint": "amd/chatglm3-6b-onnx-ryzenai-npu",
183
198
  "recipe": "oga-npu",
184
199
  "suggested": false,
185
- "size": 3.53
186
- },
187
- "AMD-OLMo-1B-Instruct-NPU": {
188
- "checkpoint": "amd/AMD-OLMo-1B-SFT-DPO-awq-g128-int4-asym-bf16-onnx-ryzen-strix",
189
- "recipe": "oga-npu",
190
- "suggested": false,
191
- "size": 2.56
200
+ "size": 7.03
192
201
  },
193
202
  "Llama-3.2-1B-Instruct-DirectML": {
194
203
  "checkpoint": "amd/Llama-3.2-1B-Instruct-dml-int4-awq-block-128-directml",
@@ -411,13 +420,33 @@
411
420
  "labels": ["coding"],
412
421
  "size": 4.68
413
422
  },
414
- "Gemma-3-4b-it-FLM": {
423
+ "gpt-oss-20b-FLM": {
424
+ "checkpoint": "gpt-oss:20b",
425
+ "recipe": "flm",
426
+ "suggested": true,
427
+ "labels": ["reasoning"],
428
+ "size": 13.4
429
+ },
430
+ "Gemma3-1b-it-FLM": {
431
+ "checkpoint": "gemma3:1b",
432
+ "recipe": "flm",
433
+ "suggested": true,
434
+ "size": 1.17
435
+ },
436
+ "Gemma3-4b-it-FLM": {
415
437
  "checkpoint": "gemma3:4b",
416
438
  "recipe": "flm",
417
439
  "suggested": true,
418
440
  "labels": ["hot","vision"],
419
441
  "size": 5.26
420
442
  },
443
+ "Qwen3-0.6b-FLM": {
444
+ "checkpoint": "qwen3:0.6b",
445
+ "recipe": "flm",
446
+ "suggested": true,
447
+ "labels": ["reasoning"],
448
+ "size": 0.66
449
+ },
421
450
  "Qwen3-4B-Instruct-2507-FLM": {
422
451
  "checkpoint": "qwen3-it:4b",
423
452
  "recipe": "flm",
@@ -432,6 +461,12 @@
432
461
  "labels": ["reasoning"],
433
462
  "size": 5.57
434
463
  },
464
+ "Llama-3.1-8B-FLM": {
465
+ "checkpoint": "llama3.1:8b",
466
+ "recipe": "flm",
467
+ "suggested": true,
468
+ "size": 5.36
469
+ },
435
470
  "Llama-3.2-1B-FLM": {
436
471
  "checkpoint": "llama3.2:1b",
437
472
  "recipe": "flm",
@@ -443,17 +478,5 @@
443
478
  "recipe": "flm",
444
479
  "suggested": true,
445
480
  "size": 2.62
446
- },
447
- "Llama-3.1-8B-FLM": {
448
- "checkpoint": "llama3.1:8b",
449
- "recipe": "flm",
450
- "suggested": true,
451
- "size": 5.36
452
- },
453
- "gpt-oss-20b-FLM": {
454
- "checkpoint": "gpt-oss:20b",
455
- "recipe": "flm",
456
- "suggested": true,
457
- "size": 13.4
458
481
  }
459
482
  }