See axolotl config

axolotl version: 0.3.0

base_model: codellama/CodeLlama-7b-hf
base_model_config: codellama/CodeLlama-7b-hf
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
is_llama_derived_model: true
hub_model_id: EvolCodeLlama-7b

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: mlabonne/Evol-Instruct-Python-1k
    type: alpaca
dataset_prepared_path: last_run_prepared
val_set_size: 0.02
output_dir: ./qlora-out

adapter: qlora
lora_model_dir:

sequence_len: 2048
sample_packing: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: axolotl
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 3
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 100
eval_steps: 0.01
save_strategy: epoch
save_steps:
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

EvolCodeLlama-7b

This model is a fine-tuned version of codellama/CodeLlama-7b-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3797

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• load_in_8bit: False
• load_in_4bit: True
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: nf4
• bnb_4bit_use_double_quant: True
• bnb_4bit_compute_dtype: bfloat16

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
0.3472	0.01	1	0.4986
0.3139	0.03	4	0.4985
0.2981	0.07	8	0.4983
0.4311	0.1	12	0.4979
0.3958	0.14	16	0.4960
0.335	0.17	20	0.4915
0.4286	0.2	24	0.4808
0.4011	0.24	28	0.4629
0.3269	0.27	32	0.4445
0.2559	0.31	36	0.4284
0.3786	0.34	40	0.4174
0.2967	0.37	44	0.4107
0.2677	0.41	48	0.4027
0.2455	0.44	52	0.3959
0.3267	0.47	56	0.3916
0.2902	0.51	60	0.3882
0.1845	0.54	64	0.3878
0.2593	0.58	68	0.3869
0.3104	0.61	72	0.3836
0.3799	0.64	76	0.3819
0.2059	0.68	80	0.3794
0.3177	0.71	84	0.3792
0.2307	0.75	88	0.3768
0.282	0.78	92	0.3749
0.2713	0.81	96	0.3738
0.2948	0.85	100	0.3725
0.2311	0.88	104	0.3713
0.2516	0.92	108	0.3716
0.2462	0.95	112	0.3715
0.2035	0.98	116	0.3711
0.2638	1.02	120	0.3712
0.2477	1.05	124	0.3726
0.1986	1.08	128	0.3682
0.2292	1.12	132	0.3671
0.1549	1.15	136	0.3680
0.1953	1.19	140	0.3683
0.224	1.22	144	0.3671
0.1941	1.25	148	0.3687
0.2234	1.29	152	0.3709
0.2659	1.32	156	0.3700
0.2535	1.36	160	0.3689
0.2115	1.39	164	0.3683
0.2481	1.42	168	0.3693
0.2101	1.46	172	0.3699
0.228	1.49	176	0.3697
0.3159	1.53	180	0.3680
0.2257	1.56	184	0.3664
0.1684	1.59	188	0.3670
0.2277	1.63	192	0.3663
0.2787	1.66	196	0.3668
0.2284	1.69	200	0.3654
0.2789	1.73	204	0.3640
0.2089	1.76	208	0.3632
0.3387	1.8	212	0.3633
0.2677	1.83	216	0.3610
0.2684	1.86	220	0.3609
0.2458	1.9	224	0.3610
0.2808	1.93	228	0.3602
0.2895	1.97	232	0.3596
0.323	2.0	236	0.3591
0.2105	2.03	240	0.3623
0.1911	2.07	244	0.3720
0.2888	2.1	248	0.3802
0.1958	2.13	252	0.3748
0.1785	2.17	256	0.3701
0.2604	2.2	260	0.3709
0.2212	2.24	264	0.3737
0.1996	2.27	268	0.3772
0.1567	2.3	272	0.3778
0.1777	2.34	276	0.3778
0.2642	2.37	280	0.3785
0.1907	2.4	284	0.3796
0.1637	2.44	288	0.3785
0.1778	2.47	292	0.3785
0.144	2.51	296	0.3789
0.1758	2.54	300	0.3788
0.2018	2.57	304	0.3784
0.3126	2.61	308	0.3783
0.1623	2.64	312	0.3790
0.223	2.68	316	0.3798
0.2109	2.71	320	0.3797
0.1606	2.74	324	0.3797
0.2226	2.78	328	0.3796
0.2068	2.81	332	0.3798
0.1547	2.85	336	0.3797
0.2513	2.88	340	0.3796
0.2688	2.91	344	0.3797
0.1481	2.95	348	0.3796
0.1443	2.98	352	0.3797

Framework versions

• PEFT 0.7.0
• Transformers 4.37.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.16.1
• Tokenizers 0.15.0