intervitens/internlm2-limarp-lora

Don't use this yet, there's a problem with the llamafied internlm2 tokenizer. Prompt format: ChatML.

See axolotl config

axolotl version: 0.3.0

base_model: /data/internlm2-base-20b-llama
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
is_llama_derived_model: true

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: /data/train-all-8k.jsonl
    type: completion
dataset_prepared_path:
val_set_size: 0.05
output_dir: /data/internlm-limarp-lora-out

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 128
lora_alpha: 64
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.00002

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

gradient_checkpointing: true
gradient_checkpointing_kwargs: 
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

internlm-limarp-lora

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.1216

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: True
• load_in_4bit: False
• 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: fp4
• bnb_4bit_use_double_quant: False
• bnb_4bit_compute_dtype: float32

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• 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: 10
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
2.3563	0.01	1	2.3995
2.1815	0.25	37	2.2693
2.1364	0.51	74	2.1684
2.1355	0.76	111	2.1526
2.1624	1.03	148	2.1435
2.1326	1.28	185	2.1367
1.9987	1.54	222	2.1330
2.0494	1.79	259	2.1291
2.0505	2.04	296	2.1266
2.075	2.3	333	2.1243
2.0183	2.55	370	2.1229
2.1047	2.81	407	2.1227
2.1309	3.06	444	2.1218
2.1249	3.31	481	2.1214
2.1423	3.57	518	2.1214
2.0913	3.82	555	2.1216

Framework versions

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