yentinglin commited on
Commit
3b2ce30
·
verified ·
1 Parent(s): 08071f6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +73 -1
README.md CHANGED
@@ -260,4 +260,76 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
260
 
261
  ## Model Card Contact
262
 
263
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
260
 
261
  ## Model Card Contact
262
 
263
+ [More Information Needed]
264
+
265
+
266
+ [<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
267
+ <details><summary>See axolotl config</summary>
268
+
269
+ axolotl version: [`a98526ef7843a3e8aa006f260e6b4fb8912b5f1a`](https://github.com/axolotl-ai-cloud/axolotl/tree/a98526ef7843a3e8aa006f260e6b4fb8912b5f1a)
270
+
271
+ ```yaml
272
+ base_model: mistralai/Mistral-Small-24B-Instruct-2501
273
+
274
+ plugins:
275
+ - axolotl.integrations.liger.LigerPlugin
276
+ liger_rope: true
277
+ liger_rms_norm: true
278
+ liger_swiglu: true
279
+ liger_fused_linear_cross_entropy: true
280
+
281
+ datasets:
282
+ - path: yentinglin/s1K-1.1-trl-format
283
+ type: chat_template
284
+ chat_template: tokenizer_default
285
+ field_messages: messages
286
+ message_field_role: role
287
+ message_field_content: content
288
+ - path: open-r1/OpenR1-Math-220k
289
+ type: chat_template
290
+ chat_template: tokenizer_default
291
+ field_messages: messages
292
+ message_field_role: from
293
+ message_field_content: value
294
+ dataset_prepared_path:
295
+ val_set_size: 0.0
296
+ output_dir: ./placeholder/
297
+
298
+ sequence_len: 32768
299
+ sample_packing: true
300
+ eval_sample_packing: False
301
+ pad_to_sequence_len: true
302
+
303
+ wandb_project: Reasoning
304
+ wandb_entity:
305
+ wandb_watch:
306
+ wandb_name: Mistral-24B-SFT-220k
307
+ wandb_log_model:
308
+
309
+ gradient_accumulation_steps: 4
310
+ micro_batch_size: 1
311
+ num_epochs: 5
312
+ optimizer: adamw_torch_fused
313
+ lr_scheduler: cosine
314
+ learning_rate: 2e-5
315
+
316
+ train_on_inputs: false
317
+ group_by_length: false
318
+ bf16: auto
319
+ tf32: false
320
+
321
+ gradient_checkpointing: true
322
+ gradient_checkpointing_kwargs:
323
+ use_reentrant: false
324
+ logging_steps: 1
325
+ flash_attention: true
326
+
327
+ warmup_ratio: 0.1
328
+ saves_per_epoch: 2
329
+ weight_decay: 0.0
330
+ deepspeed: deepspeed_configs/zero3_bf16.json
331
+ special_tokens:
332
+ pad_token: "<pad>"
333
+ ```
334
+
335
+ </details><br>