Views
No views yet
0.8.0.dev01base_model: meta-llama/Llama-3.1-70B
2# Automatically upload checkpoint and final model to HF
3# hub_model_id: username/custom_model_name
4#
5
6plugins:
7 - axolotl.integrations.liger.LigerPlugin
8 - axolotl.integrations.spectrum.SpectrumPlugin
9spectrum_top_fraction: 0.5
10spectrum_model_name: meta-llama/Meta-Llama-3.1-70B
11liger_rope: true
12liger_rms_norm: true
13liger_glu_activation: true
14liger_fused_linear_cross_entropy: true
15
16strict: false
17
18chat_template: llama3
19datasets:
20 - path: bespokelabs/Bespoke-Stratos-17k
21 field_messages: conversations
22 message_property_mappings:
23 content: value
24 role: from
25 split: train
26 type: chat_template
27dataset_prepared_path: last_run_prepared
28val_set_size: 0.0
29output_dir: ./outputs/out/reasoning-70b-stratos
30save_safetensors: true
31
32wandb_project: reasoning-70b-stratos
33wandb_entity: axolotl-ai
34wandb_watch:
35wandb_name:
36wandb_log_model:
37
38sequence_len: 16384
39sample_packing: true
40pad_to_sequence_len: true
41
42gradient_accumulation_steps: 1
43micro_batch_size: 4
44num_epochs: 3
45optimizer: adamw_torch_fused
46lr_scheduler: rex
47learning_rate: 5.0e-6
48embedding_lr: 1.0e-5
49max_grad_norm: 1.0
50
51train_on_inputs: false
52group_by_length: false
53bf16: true
54tf32: true
55
56gradient_checkpointing: offload
57gradient_checkpointing_kwargs:
58 use_reentrant: true
59logging_steps: 1
60flash_attention: true
61
62warmup_steps: 20
63evals_per_epoch: 4
64saves_per_epoch: 1
65weight_decay: 0.01
66deepspeed: deepspeed_configs/zero3_bf16_cpuoffload_params.json
67special_tokens:
68 pad_token: <|finetune_right_pad_id|>
69 eos_token: <|eot_id|>
70added_tokens_overrides:
71 128011: <think>
72 128012: </think>
73 128013: <|begin_of_thought|>
74 128014: <|end_of_thought|>
75 128015: <|begin_of_solution|>
76 128016: <|end_of_solution|>
77fix_untrained_tokens:
78 - 128011
79 - 128012
80 - 128013
81 - 128014
82 - 128015
83 - 128016
84