metadata
base_model: habanoz/TinyLlama-1.1B-2T-lr-2e-4-3ep-dolly-15k-instruct-v1
datasets:
- databricks/databricks-dolly-15k
inference: false
language:
- en
license: apache-2.0
model_creator: habanoz
model_name: TinyLlama-1.1B-2T-lr-2e-4-3ep-dolly-15k-instruct-v1
pipeline_tag: text-generation
quantized_by: afrideva
tags:
- gguf
- ggml
- quantized
- q2_k
- q3_k_m
- q4_k_m
- q5_k_m
- q6_k
- q8_0
habanoz/TinyLlama-1.1B-2T-lr-2e-4-3ep-dolly-15k-instruct-v1-GGUF
Quantized GGUF model files for TinyLlama-1.1B-2T-lr-2e-4-3ep-dolly-15k-instruct-v1 from habanoz
| Name | Quant method | Size |
|---|---|---|
| tinyllama-1.1b-2t-lr-2e-4-3ep-dolly-15k-instruct-v1.fp16.gguf | fp16 | 2.20 GB |
| tinyllama-1.1b-2t-lr-2e-4-3ep-dolly-15k-instruct-v1.q2_k.gguf | q2_k | 483.12 MB |
| tinyllama-1.1b-2t-lr-2e-4-3ep-dolly-15k-instruct-v1.q3_k_m.gguf | q3_k_m | 550.82 MB |
| tinyllama-1.1b-2t-lr-2e-4-3ep-dolly-15k-instruct-v1.q4_k_m.gguf | q4_k_m | 668.79 MB |
| tinyllama-1.1b-2t-lr-2e-4-3ep-dolly-15k-instruct-v1.q5_k_m.gguf | q5_k_m | 783.02 MB |
| tinyllama-1.1b-2t-lr-2e-4-3ep-dolly-15k-instruct-v1.q6_k.gguf | q6_k | 904.39 MB |
| tinyllama-1.1b-2t-lr-2e-4-3ep-dolly-15k-instruct-v1.q8_0.gguf | q8_0 | 1.17 GB |
Original Model Card:
TinyLlama/TinyLlama-1.1B-intermediate-step-955k-token-2T finetuned using dolly dataset.
Training took 1 hour on an 'ml.g5.xlarge' instance.
hyperparameters ={
'num_train_epochs': 3, # number of training epochs
'per_device_train_batch_size': 6, # batch size for training
'gradient_accumulation_steps': 2, # Number of updates steps to accumulate
'gradient_checkpointing': True, # save memory but slower backward pass
'bf16': True, # use bfloat16 precision
'tf32': True, # use tf32 precision
'learning_rate': 2e-4, # learning rate
'max_grad_norm': 0.3, # Maximum norm (for gradient clipping)
'warmup_ratio': 0.03, # warmup ratio
"lr_scheduler_type":"constant", # learning rate scheduler
'save_strategy': "epoch", # save strategy for checkpoints
"logging_steps": 10, # log every x steps
'merge_adapters': True, # wether to merge LoRA into the model (needs more memory)
'use_flash_attn': True, # Whether to use Flash Attention
}