Downloads · 30 days
32
2% of all-time downloads
CultriX_-_Lama-DPOlphin-8B-gguf
RichardErkhov/CultriX_-_Lama-DPOlphin-8B-gguf
CultriX_-_Lama-DPOlphin-8B-gguf is a machine learning model from RichardErkhov. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
All-time downloads
1.4K
Public
Repo size
92.9 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.gguf108 GB · 100%
At a glance
- Access
- Public
- Created
- Sep 22, 2024
- Updated
- Sep 22, 2024
- SHA
- bf4d1dfd
Model card
From the Hugging Face model README
Quantization made by Richard Erkhov.
Lama-DPOlphin-8B - GGUF
- Model creator: https://huggingface.co/CultriX/
- Original model: https://huggingface.co/CultriX/Lama-DPOlphin-8B/
| Name | Quant method | Size |
|---|---|---|
| Lama-DPOlphin-8B.Q2_K.gguf | Q2_K | 2.96GB |
| Lama-DPOlphin-8B.IQ3_XS.gguf | IQ3_XS | 3.28GB |
| Lama-DPOlphin-8B.IQ3_S.gguf | IQ3_S | 3.43GB |
| Lama-DPOlphin-8B.Q3_K_S.gguf | Q3_K_S | 3.41GB |
| Lama-DPOlphin-8B.IQ3_M.gguf | IQ3_M | 3.52GB |
| Lama-DPOlphin-8B.Q3_K.gguf | Q3_K | 3.74GB |
| Lama-DPOlphin-8B.Q3_K_M.gguf | Q3_K_M | 3.74GB |
| Lama-DPOlphin-8B.Q3_K_L.gguf | Q3_K_L | 4.03GB |
| Lama-DPOlphin-8B.IQ4_XS.gguf | IQ4_XS | 4.18GB |
| Lama-DPOlphin-8B.Q4_0.gguf | Q4_0 | 4.34GB |
| Lama-DPOlphin-8B.IQ4_NL.gguf | IQ4_NL | 4.38GB |
| Lama-DPOlphin-8B.Q4_K_S.gguf | Q4_K_S | 4.37GB |
| Lama-DPOlphin-8B.Q4_K.gguf | Q4_K | 4.58GB |
| Lama-DPOlphin-8B.Q4_K_M.gguf | Q4_K_M | 4.58GB |
| Lama-DPOlphin-8B.Q4_1.gguf | Q4_1 | 4.78GB |
| Lama-DPOlphin-8B.Q5_0.gguf | Q5_0 | 5.21GB |
| Lama-DPOlphin-8B.Q5_K_S.gguf | Q5_K_S | 5.21GB |
| Lama-DPOlphin-8B.Q5_K.gguf | Q5_K | 5.34GB |
| Lama-DPOlphin-8B.Q5_K_M.gguf | Q5_K_M | 5.34GB |
| Lama-DPOlphin-8B.Q5_1.gguf | Q5_1 | 5.65GB |
| Lama-DPOlphin-8B.Q6_K.gguf | Q6_K | 6.14GB |
| Lama-DPOlphin-8B.Q8_0.gguf | Q8_0 | 7.95GB |
Original model description:
license: apache-2.0 datasets:
- mlabonne/orpo-dpo-mix-40k-flat language:
- en base_model: cognitivecomputations/dolphin-2.9.4-llama3.1-8b pipeline_tag: text-generation tags:
- dpo
- axolotl library_name: transformers
Axolotl configuration:
base_model: cognitivecomputations/dolphin-2.9.4-llama3.1-8b
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer
tokenizer:
name_or_path: "https://huggingface.co/cognitivecomputations/dolphin-2.9.4-llama3.1-8b/resolve/main/tokenizer.json"
load_in_8bit: false
load_in_4bit: true
strict: false
save_safetensors: true
bnb_4bit_quant_type: "nf4"
bnb_4bit_compute_dtype: "bf16"
bnb_4bit_use_double_quant: true
rl: dpo
chat_template: chatml
datasets:
- path: mlabonne/orpo-dpo-mix-40k-flat
split: train
type: chatml.intel
dataset_prepared_path: /workspace/axolotl/dataset-prepared
val_set_size: 0.0
output_dir: ./out
adapter: qlora
lora_model_dir:
sequence_len: 2048
sample_packing: false
pad_to_sequence_len: false
lora_r: 64
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
wandb_project: axolotl
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
gradient_accumulation_steps: 4 # Reduced from 8 to 4 due to large VRAM
micro_batch_size: 2 # Increased micro-batch size to 2
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 5e-6
train_on_inputs: false
group_by_length: false
bf16: true # Use bf16 as it is optimal for A40 GPUs
fp16: false
tf32: true # TF32 is supported by A40 and improves performance
gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
warmup_steps: 100
evals_per_epoch: 0
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed: deepspeed_configs/zero2.json # Enable DeepSpeed with ZeRO Stage 2
weight_decay: 0.0
special_tokens:
pad_token: <|end_of_text|>
- Created
- Sep 22, 2024
- Updated
- Sep 22, 2024