code380 commited on
Commit
4454f79
·
verified ·
1 Parent(s): 9e31b0f

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +445 -0
README.md ADDED
@@ -0,0 +1,445 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: OpenLLM-Ro/RoLlama3-8b-Instruct-DPO
3
+ datasets:
4
+ - OpenLLM-Ro/ro_dpo_helpsteer
5
+ language:
6
+ - ro
7
+ license: cc-by-nc-4.0
8
+ tags:
9
+ - llama-cpp
10
+ - gguf-my-repo
11
+ model-index:
12
+ - name: OpenLLM-Ro/RoLlama3-8b-Instruct-DPO-2024-10-09
13
+ results:
14
+ - task:
15
+ type: text-generation
16
+ dataset:
17
+ name: RoMT-Bench
18
+ type: RoMT-Bench
19
+ metrics:
20
+ - type: Score
21
+ value: 5.87
22
+ name: Score
23
+ - type: Score
24
+ value: 6.22
25
+ name: First turn
26
+ - type: Score
27
+ value: 5.49
28
+ name: Second turn
29
+ - task:
30
+ type: text-generation
31
+ dataset:
32
+ name: RoCulturaBench
33
+ type: RoCulturaBench
34
+ metrics:
35
+ - type: Score
36
+ value: 4.4
37
+ name: Score
38
+ - task:
39
+ type: text-generation
40
+ dataset:
41
+ name: Romanian_Academic_Benchmarks
42
+ type: Romanian_Academic_Benchmarks
43
+ metrics:
44
+ - type: accuracy
45
+ value: 49.96
46
+ name: Average accuracy
47
+ - task:
48
+ type: text-generation
49
+ dataset:
50
+ name: OpenLLM-Ro/ro_arc_challenge
51
+ type: OpenLLM-Ro/ro_arc_challenge
52
+ metrics:
53
+ - type: accuracy
54
+ value: 46.29
55
+ name: Average accuracy
56
+ - type: accuracy
57
+ value: 44.56
58
+ name: 0-shot
59
+ - type: accuracy
60
+ value: 45.42
61
+ name: 1-shot
62
+ - type: accuracy
63
+ value: 46.1
64
+ name: 3-shot
65
+ - type: accuracy
66
+ value: 46.27
67
+ name: 5-shot
68
+ - type: accuracy
69
+ value: 46.96
70
+ name: 10-shot
71
+ - type: accuracy
72
+ value: 48.41
73
+ name: 25-shot
74
+ - task:
75
+ type: text-generation
76
+ dataset:
77
+ name: OpenLLM-Ro/ro_mmlu
78
+ type: OpenLLM-Ro/ro_mmlu
79
+ metrics:
80
+ - type: accuracy
81
+ value: 53.29
82
+ name: Average accuracy
83
+ - type: accuracy
84
+ value: 52.33
85
+ name: 0-shot
86
+ - type: accuracy
87
+ value: 52.86
88
+ name: 1-shot
89
+ - type: accuracy
90
+ value: 54.06
91
+ name: 3-shot
92
+ - type: accuracy
93
+ value: 53.9
94
+ name: 5-shot
95
+ - task:
96
+ type: text-generation
97
+ dataset:
98
+ name: OpenLLM-Ro/ro_winogrande
99
+ type: OpenLLM-Ro/ro_winogrande
100
+ metrics:
101
+ - type: accuracy
102
+ value: 65.57
103
+ name: Average accuracy
104
+ - type: accuracy
105
+ value: 64.33
106
+ name: 0-shot
107
+ - type: accuracy
108
+ value: 64.72
109
+ name: 1-shot
110
+ - type: accuracy
111
+ value: 66.3
112
+ name: 3-shot
113
+ - type: accuracy
114
+ value: 66.93
115
+ name: 5-shot
116
+ - task:
117
+ type: text-generation
118
+ dataset:
119
+ name: OpenLLM-Ro/ro_hellaswag
120
+ type: OpenLLM-Ro/ro_hellaswag
121
+ metrics:
122
+ - type: accuracy
123
+ value: 58.15
124
+ name: Average accuracy
125
+ - type: accuracy
126
+ value: 57.45
127
+ name: 0-shot
128
+ - type: accuracy
129
+ value: 57.65
130
+ name: 1-shot
131
+ - type: accuracy
132
+ value: 58.18
133
+ name: 3-shot
134
+ - type: accuracy
135
+ value: 58.64
136
+ name: 5-shot
137
+ - type: accuracy
138
+ value: 58.85
139
+ name: 10-shot
140
+ - task:
141
+ type: text-generation
142
+ dataset:
143
+ name: OpenLLM-Ro/ro_gsm8k
144
+ type: OpenLLM-Ro/ro_gsm8k
145
+ metrics:
146
+ - type: accuracy
147
+ value: 34.77
148
+ name: Average accuracy
149
+ - type: accuracy
150
+ value: 32.52
151
+ name: 1-shot
152
+ - type: accuracy
153
+ value: 33.97
154
+ name: 3-shot
155
+ - type: accuracy
156
+ value: 37.83
157
+ name: 5-shot
158
+ - task:
159
+ type: text-generation
160
+ dataset:
161
+ name: OpenLLM-Ro/ro_truthfulqa
162
+ type: OpenLLM-Ro/ro_truthfulqa
163
+ metrics:
164
+ - type: accuracy
165
+ value: 41.7
166
+ name: Average accuracy
167
+ - task:
168
+ type: text-generation
169
+ dataset:
170
+ name: LaRoSeDa_binary
171
+ type: LaRoSeDa_binary
172
+ metrics:
173
+ - type: macro-f1
174
+ value: 97.48
175
+ name: Average macro-f1
176
+ - type: macro-f1
177
+ value: 97.67
178
+ name: 0-shot
179
+ - type: macro-f1
180
+ value: 97.07
181
+ name: 1-shot
182
+ - type: macro-f1
183
+ value: 97.4
184
+ name: 3-shot
185
+ - type: macro-f1
186
+ value: 97.8
187
+ name: 5-shot
188
+ - task:
189
+ type: text-generation
190
+ dataset:
191
+ name: LaRoSeDa_multiclass
192
+ type: LaRoSeDa_multiclass
193
+ metrics:
194
+ - type: macro-f1
195
+ value: 54.0
196
+ name: Average macro-f1
197
+ - type: macro-f1
198
+ value: 58.49
199
+ name: 0-shot
200
+ - type: macro-f1
201
+ value: 55.93
202
+ name: 1-shot
203
+ - type: macro-f1
204
+ value: 47.7
205
+ name: 3-shot
206
+ - type: macro-f1
207
+ value: 53.89
208
+ name: 5-shot
209
+ - task:
210
+ type: text-generation
211
+ dataset:
212
+ name: LaRoSeDa_binary_finetuned
213
+ type: LaRoSeDa_binary_finetuned
214
+ metrics:
215
+ - type: macro-f1
216
+ value: 0.0
217
+ name: Average macro-f1
218
+ - task:
219
+ type: text-generation
220
+ dataset:
221
+ name: LaRoSeDa_multiclass_finetuned
222
+ type: LaRoSeDa_multiclass_finetuned
223
+ metrics:
224
+ - type: macro-f1
225
+ value: 0.0
226
+ name: Average macro-f1
227
+ - task:
228
+ type: text-generation
229
+ dataset:
230
+ name: WMT_EN-RO
231
+ type: WMT_EN-RO
232
+ metrics:
233
+ - type: bleu
234
+ value: 22.09
235
+ name: Average bleu
236
+ - type: bleu
237
+ value: 8.63
238
+ name: 0-shot
239
+ - type: bleu
240
+ value: 25.89
241
+ name: 1-shot
242
+ - type: bleu
243
+ value: 26.79
244
+ name: 3-shot
245
+ - type: bleu
246
+ value: 27.05
247
+ name: 5-shot
248
+ - task:
249
+ type: text-generation
250
+ dataset:
251
+ name: WMT_RO-EN
252
+ type: WMT_RO-EN
253
+ metrics:
254
+ - type: bleu
255
+ value: 23.0
256
+ name: Average bleu
257
+ - type: bleu
258
+ value: 3.56
259
+ name: 0-shot
260
+ - type: bleu
261
+ value: 20.66
262
+ name: 1-shot
263
+ - type: bleu
264
+ value: 33.56
265
+ name: 3-shot
266
+ - type: bleu
267
+ value: 34.22
268
+ name: 5-shot
269
+ - task:
270
+ type: text-generation
271
+ dataset:
272
+ name: WMT_EN-RO_finetuned
273
+ type: WMT_EN-RO_finetuned
274
+ metrics:
275
+ - type: bleu
276
+ value: 0.0
277
+ name: Average bleu
278
+ - task:
279
+ type: text-generation
280
+ dataset:
281
+ name: WMT_RO-EN_finetuned
282
+ type: WMT_RO-EN_finetuned
283
+ metrics:
284
+ - type: bleu
285
+ value: 0.0
286
+ name: Average bleu
287
+ - task:
288
+ type: text-generation
289
+ dataset:
290
+ name: XQuAD
291
+ type: XQuAD
292
+ metrics:
293
+ - type: exact_match
294
+ value: 26.05
295
+ name: Average exact_match
296
+ - type: f1
297
+ value: 42.77
298
+ name: Average f1
299
+ - task:
300
+ type: text-generation
301
+ dataset:
302
+ name: XQuAD_finetuned
303
+ type: XQuAD_finetuned
304
+ metrics:
305
+ - type: exact_match
306
+ value: 0.0
307
+ name: Average exact_match
308
+ - type: f1
309
+ value: 0.0
310
+ name: Average f1
311
+ - task:
312
+ type: text-generation
313
+ dataset:
314
+ name: STS
315
+ type: STS
316
+ metrics:
317
+ - type: spearman
318
+ value: 79.64
319
+ name: Average spearman
320
+ - type: pearson
321
+ value: 79.52
322
+ name: Average pearson
323
+ - task:
324
+ type: text-generation
325
+ dataset:
326
+ name: STS_finetuned
327
+ type: STS_finetuned
328
+ metrics:
329
+ - type: spearman
330
+ value: 0.0
331
+ name: Average spearman
332
+ - type: pearson
333
+ value: 0.0
334
+ name: Average pearson
335
+ - task:
336
+ type: text-generation
337
+ dataset:
338
+ name: XQuAD_EM
339
+ type: XQuAD_EM
340
+ metrics:
341
+ - type: exact_match
342
+ value: 11.26
343
+ name: 0-shot
344
+ - type: exact_match
345
+ value: 34.29
346
+ name: 1-shot
347
+ - type: exact_match
348
+ value: 29.24
349
+ name: 3-shot
350
+ - type: exact_match
351
+ value: 29.41
352
+ name: 5-shot
353
+ - task:
354
+ type: text-generation
355
+ dataset:
356
+ name: XQuAD_F1
357
+ type: XQuAD_F1
358
+ metrics:
359
+ - type: f1
360
+ value: 22.98
361
+ name: 0-shot
362
+ - type: f1
363
+ value: 54.48
364
+ name: 1-shot
365
+ - type: f1
366
+ value: 46.18
367
+ name: 3-shot
368
+ - type: f1
369
+ value: 47.43
370
+ name: 5-shot
371
+ - task:
372
+ type: text-generation
373
+ dataset:
374
+ name: STS_Spearman
375
+ type: STS_Spearman
376
+ metrics:
377
+ - type: spearman
378
+ value: 79.99
379
+ name: 1-shot
380
+ - type: spearman
381
+ value: 78.42
382
+ name: 3-shot
383
+ - type: spearman
384
+ value: 80.51
385
+ name: 5-shot
386
+ - task:
387
+ type: text-generation
388
+ dataset:
389
+ name: STS_Pearson
390
+ type: STS_Pearson
391
+ metrics:
392
+ - type: pearson
393
+ value: 80.59
394
+ name: 1-shot
395
+ - type: pearson
396
+ value: 78.11
397
+ name: 3-shot
398
+ - type: pearson
399
+ value: 79.87
400
+ name: 5-shot
401
+ ---
402
+
403
+ # code380/RoLlama3-8b-Instruct-DPO-Q8_0-GGUF
404
+ This model was converted to GGUF format from [`OpenLLM-Ro/RoLlama3-8b-Instruct-DPO`](https://huggingface.co/OpenLLM-Ro/RoLlama3-8b-Instruct-DPO) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space.
405
+ Refer to the [original model card](https://huggingface.co/OpenLLM-Ro/RoLlama3-8b-Instruct-DPO) for more details on the model.
406
+
407
+ ## Use with llama.cpp
408
+ Install llama.cpp through brew (works on Mac and Linux)
409
+
410
+ ```bash
411
+ brew install llama.cpp
412
+
413
+ ```
414
+ Invoke the llama.cpp server or the CLI.
415
+
416
+ ### CLI:
417
+ ```bash
418
+ llama-cli --hf-repo code380/RoLlama3-8b-Instruct-DPO-Q8_0-GGUF --hf-file rollama3-8b-instruct-dpo-q8_0.gguf -p "The meaning to life and the universe is"
419
+ ```
420
+
421
+ ### Server:
422
+ ```bash
423
+ llama-server --hf-repo code380/RoLlama3-8b-Instruct-DPO-Q8_0-GGUF --hf-file rollama3-8b-instruct-dpo-q8_0.gguf -c 2048
424
+ ```
425
+
426
+ Note: You can also use this checkpoint directly through the [usage steps](https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#usage) listed in the Llama.cpp repo as well.
427
+
428
+ Step 1: Clone llama.cpp from GitHub.
429
+ ```
430
+ git clone https://github.com/ggerganov/llama.cpp
431
+ ```
432
+
433
+ Step 2: Move into the llama.cpp folder and build it with `LLAMA_CURL=1` flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).
434
+ ```
435
+ cd llama.cpp && LLAMA_CURL=1 make
436
+ ```
437
+
438
+ Step 3: Run inference through the main binary.
439
+ ```
440
+ ./llama-cli --hf-repo code380/RoLlama3-8b-Instruct-DPO-Q8_0-GGUF --hf-file rollama3-8b-instruct-dpo-q8_0.gguf -p "The meaning to life and the universe is"
441
+ ```
442
+ or
443
+ ```
444
+ ./llama-server --hf-repo code380/RoLlama3-8b-Instruct-DPO-Q8_0-GGUF --hf-file rollama3-8b-instruct-dpo-q8_0.gguf -c 2048
445
+ ```