1base_model: meta-llama/CodeLlama-7b-hf
2bf16: auto
3dataset_processes: 32
4datasets:
5- message_property_mappings:
6 content: content
7 role: role
8 path: steve329/svgen-500k-2k
9 trust_remote_code: false
10 type: alpaca
11gradient_accumulation_steps: 1
12gradient_checkpointing: false
13learning_rate: 0.0002
14lisa_layers_attribute: model.layers
15load_best_model_at_end: false
16load_in_4bit: false
17load_in_8bit: false
18lora_alpha: 16
19lora_dropout: 0.05
20lora_r: 8
21lora_target_modules:
22- q_proj
23- v_proj
24- k_proj
25- o_proj
26- gate_proj
27- down_proj
28- up_proj
29loraplus_lr_embedding: 1.0e-06
30lr_scheduler: cosine
31max_prompt_len: 1400
32mean_resizing_embeddings: false
33micro_batch_size: 1
34num_epochs: 1.0
35optimizer: adamw_bnb_8bit
36output_dir: ./outputs/mymodel
37pretrain_multipack_attn: true
38pretrain_multipack_buffer_size: 10000
39qlora_sharded_model_loading: false
40ray_num_workers: 1
41resources_per_worker:
42 GPU: 1
43sample_packing_bin_size: 200
44sample_packing_group_size: 100000
45save_only_model: false
46save_safetensors: true
47sequence_len: 1400
48shuffle_merged_datasets: true
49skip_prepare_dataset: false
50strict: false
51train_on_inputs: false
52trl:
53 log_completions: false
54 ref_model_mixup_alpha: 0.9
55 ref_model_sync_steps: 64
56 sync_ref_model: false
57 use_vllm: false
58 vllm_device: auto
59 vllm_dtype: auto
60 vllm_gpu_memory_utilization: 0.9
61use_ray: false
62val_set_size: 0.0
63weight_decay: 0.0
64
