See axolotl config

axolotl version: 0.4.0

base_model: mistralai/Mistral-7B-v0.1
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_llama_derived_model: true
hub_model_id: NistCodeLlama-7b
sample_packing: false
eval_sample_packing: false
load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: rkreddyp/nist_800_53
    ds_type: json
    type:
              field_instruction: question
              field_input: context
              field_output: answer
              format: |-
                [INST] Using the schema context below, generate a SQL query that answers the question.
                {input}
                {instruction} [/INST]   


dataset_prepared_path:
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-nist
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>"

NistCodeLlama-7b

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3414

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

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
1.4855	0.06	1	1.4808
1.4522	0.11	2	1.4811
1.4616	0.17	3	1.4788
1.5276	0.23	4	1.4746
1.4564	0.29	5	1.4662
1.4837	0.34	6	1.4515
1.4709	0.4	7	1.4280
1.3571	0.46	8	1.3903
1.4164	0.51	9	1.3363
1.3257	0.57	10	1.2692
1.2858	0.63	11	1.2027
1.2318	0.69	12	1.1364
1.1164	0.74	13	1.0595
1.0984	0.8	14	0.9748
0.9593	0.86	15	0.8923
0.8325	0.91	16	0.8137
0.8357	0.97	17	0.7426
0.6483	1.03	18	0.6868
0.7138	1.06	19	0.6400
0.6105	1.11	20	0.6027
0.6409	1.17	21	0.5686
0.5206	1.23	22	0.5317
0.521	1.29	23	0.4962
0.4409	1.34	24	0.4697
0.4678	1.4	25	0.4481
0.3731	1.46	26	0.4303
0.388	1.51	27	0.4161
0.3463	1.57	28	0.4085
0.3699	1.63	29	0.4035
0.3673	1.69	30	0.3992
0.4485	1.74	31	0.3962
0.3855	1.8	32	0.3929
0.3249	1.86	33	0.3887
0.3528	1.91	34	0.3839
0.372	1.97	35	0.3801
0.3922	2.03	36	0.3768
0.3783	2.06	37	0.3739
0.31	2.11	38	0.3721
0.275	2.17	39	0.3699
0.338	2.23	40	0.3665
0.3238	2.29	41	0.3633
0.3382	2.34	42	0.3597
0.3467	2.4	43	0.3567
0.3494	2.46	44	0.3541
0.3431	2.51	45	0.3533
0.3433	2.57	46	0.3522
0.304	2.63	47	0.3491
0.3098	2.69	48	0.3464
0.279	2.74	49	0.3443
0.3105	2.8	50	0.3425
0.2305	2.86	51	0.3414

Framework versions

• PEFT 0.8.2
• Transformers 4.38.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.17.0
• Tokenizers 0.15.0