dolphin-2.9.1-llama-3-70b-NVFP4-vllm

Dolphin 2.9.1 Llama 3 70b 🐬

Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations

Discord: https://discord.gg/cognitivecomputations

<img src="https://cdn-uploads.huggingface.co/production/uploads/63111b2d88942700629f5771/ldkN1J0WIDQwU4vutGYiD.png" width="600" />

We have retrained our LLama-3-70b fine tune to address behavioral issues in the initial 2.9 dataset. Specifically, Systemchat was causing the model to be too reliant on the system prompt. Additionally, it had an occasional quirk that would cause the model to overly reference the system prompt. We also found generation length was at times not sufficient for any given task. We identified the culprit as Ultrachat. Accounting for these concerns, we removed systemchat and ultrachat from the dataset. It is otherwise identical to dolphin-2.9.

Our appreciation for the sponsors of Dolphin 2.9.1:

• Crusoe Cloud - provided excellent on-demand 8xL40S node
• OnDemand - provided inference sponsorship

This model is based on Llama-3-70b, and is governed by META LLAMA 3 COMMUNITY LICENSE AGREEMENT

The base model has 8k context, and the full-weight fine-tuning was with 4k sequence length.

It took 3 days on an 8x H100 provided by Crusoe Cloud

This model was trained FFT on parameters selected by Laser Scanner, using ChatML prompt template format.

example:

<|im_start|>system
You are Dolphin, a helpful AI assistant.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Dolphin-2.9.1 has a variety of instruction, conversational, and coding skills. It also has initial agentic abilities and supports function calling.

Dolphin is uncensored. We have filtered the dataset to remove alignment and bias. This makes the model more compliant. You are advised to implement your own alignment layer before exposing the model as a service. It will be highly compliant with any requests, even unethical ones. Please read my blog post about uncensored models. https://erichartford.com/uncensored-models You are responsible for any content you create using this model. Enjoy responsibly.

Dolphin is licensed according to Meta's Llama license. We grant permission for any use, including commercial, that falls within accordance with Meta's Llama-3 license. Dolphin was trained on data generated from GPT4, among other models.

Evals

Training

<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

<details><summary>See axolotl config</summary>

axolotl version: 0.4.0

base_model: meta-llama/Meta-Llama-3-70B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
# load_in_4bit: true
strict: false

datasets:
  - path: /workspace/datasets/dolphin-2.9/dolphin201-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/dolphin-coder-translate-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/dolphin-coder-codegen-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/m-a-p_Code-Feedback-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/m-a-p_CodeFeedback-Filtered-Instruction-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/not_samantha_norefusals.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/Orca-Math-resort-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/agent_instruct_react_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_instruct_j1s1_3k_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_negative_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_react_10p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_tflan_cot_30p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/openhermes200k_unfiltered.jsonl
    type: sharegpt 
    conversation: chatml

chat_template: chatml
# adapter: qlora
# lora_r: 128
# lora_alpha: 16
# lora_modules_to_save: [embed_tokens, lm_head]
# lora_dropout: 0.05
# lora_target_linear: true



unfrozen_parameters:
- ^lm_head.weight$
- ^model.embed_tokens.weight$
# mlp.down_proj layers
- model.layers.40.mlp.down_proj
- model.layers.44.mlp.down_proj
- model.layers.45.mlp.down_proj
- model.layers.46.mlp.down_proj
- model.layers.43.mlp.down_proj
- model.layers.52.mlp.down_proj
- model.layers.47.mlp.down_proj
- model.layers.48.mlp.down_proj
- model.layers.39.mlp.down_proj
- model.layers.49.mlp.down_proj
- model.layers.38.mlp.down_proj
- model.layers.53.mlp.down_proj
- model.layers.41.mlp.down_proj
- model.layers.35.mlp.down_proj
- model.layers.51.mlp.down_proj
- model.layers.42.mlp.down_proj
- model.layers.37.mlp.down_proj
- model.layers.50.mlp.down_proj
- model.layers.60.mlp.down_proj
- model.layers.76.mlp.down_proj
- model.layers.54.mlp.down_proj
- model.layers.36.mlp.down_proj
- model.layers.57.mlp.down_proj
- model.layers.56.mlp.down_proj
- model.layers.55.mlp.down_proj
- model.layers.77.mlp.down_proj
- model.layers.59.mlp.down_proj
- model.layers.61.mlp.down_proj
- model.layers.58.mlp.down_proj
- model.layers.65.mlp.down_proj
- model.layers.75.mlp.down_proj
- model.layers.64.mlp.down_proj
- model.layers.62.mlp.down_proj
- model.layers.68.mlp.down_proj
- model.layers.19.mlp.down_proj
- model.layers.66.mlp.down_proj
# mlp.gate_proj layers
- model.layers.70.mlp.gate_proj
- model.layers.71.mlp.gate_proj
- model.layers.67.mlp.gate_proj
- model.layers.58.mlp.gate_proj
- model.layers.55.mlp.gate_proj
- model.layers.57.mlp.gate_proj
- model.layers.56.mlp.gate_proj
- model.layers.66.mlp.gate_proj
- model.layers.72.mlp.gate_proj
- model.layers.69.mlp.gate_proj
- model.layers.52.mlp.gate_proj
- model.layers.54.mlp.gate_proj
- model.layers.62.mlp.gate_proj
- model.layers.60.mlp.gate_proj
- model.layers.74.mlp.gate_proj
- model.layers.59.mlp.gate_proj
- model.layers.68.mlp.gate_proj
- model.layers.61.mlp.gate_proj
- model.layers.73.mlp.gate_proj
- model.layers.53.mlp.gate_proj
- model.layers.51.mlp.gate_proj
- model.layers.63.mlp.gate_proj
- model.layers.48.mlp.gate_proj
- model.layers.49.mlp.gate_proj
- model.layers.64.mlp.gate_proj
- model.layers.50.mlp.gate_proj
- model.layers.65.mlp.gate_proj
- model.layers.47.mlp.gate_proj
- model.layers.44.mlp.gate_proj
- model.layers.45.mlp.gate_proj
- model.layers.75.mlp.gate_proj
- model.layers.46.mlp.gate_proj
- model.layers.43.mlp.gate_proj
- model.layers.77.mlp.gate_proj
- model.layers.41.mlp.gate_proj
- model.layers.42.mlp.gate_proj
# mlp.up_proj layers
- model.layers.70.mlp.up_proj
- model.layers.67.mlp.up_proj
- model.layers.66.mlp.up_proj
- model.layers.69.mlp.up_proj
- model.layers.62.mlp.up_proj
- model.layers.65.mlp.up_proj
- model.layers.63.mlp.up_proj
- model.layers.68.mlp.up_proj
- model.layers.71.mlp.up_proj
- model.layers.64.mlp.up_proj
- model.layers.61.mlp.up_proj
- model.layers.58.mlp.up_proj
- model.layers.59.mlp.up_proj
- model.layers.57.mlp.up_proj
- model.layers.55.mlp.up_proj
- model.layers.72.mlp.up_proj
- model.layers.54.mlp.up_proj
- model.layers.60.mlp.up_proj
- model.layers.56.mlp.up_proj
- model.layers.73.mlp.up_proj
- model.layers.50.mlp.up_proj
- model.layers.51.mlp.up_proj
- model.layers.53.mlp.up_proj
- model.layers.74.mlp.up_proj
- model.layers.52.mlp.up_proj
- model.layers.49.mlp.up_proj
- model.layers.30.mlp.up_proj
- model.layers.34.mlp.up_proj
- model.layers.47.mlp.up_proj
- model.layers.46.mlp.up_proj
- model.layers.48.mlp.up_proj
- model.layers.38.mlp.up_proj
- model.layers.45.mlp.up_proj
- model.layers.43.mlp.up_proj
- model.layers.29.mlp.up_proj
- model.layers.42.mlp.up_proj
# self_attn.k_proj layers
- model.layers.72.self_attn.k_proj
- model.layers.75.self_attn.k_proj
- model.layers.71.self_attn.k_proj
- model.layers.74.self_attn.k_proj
- model.layers.44.self_attn.k_proj
- model.layers.31.self_attn.k_proj
- model.layers.33.self_attn.k_proj
- model.layers.34.self_attn.k_proj
- model.layers.76.self_attn.k_proj
- model.layers.78.self_attn.k_proj
- model.layers.77.self_attn.k_proj
- model.layers.22.self_attn.k_proj
- model.layers.18.self_attn.k_proj
- model.layers.60.self_attn.k_proj
- model.layers.17.self_attn.k_proj
- model.layers.56.self_attn.k_proj
- model.layers.2.self_attn.k_proj
- model.layers.21.self_attn.k_proj
- model.layers.19.self_attn.k_proj
- model.layers.23.self_attn.k_proj
- model.layers.52.self_attn.k_proj
- model.layers.35.self_attn.k_proj
- model.layers.73.self_attn.k_proj
- model.layers.15.self_attn.k_proj
- model.layers.27.self_attn.k_proj
- model.layers.29.self_attn.k_proj
- model.layers.20.self_attn.k_proj
- model.layers.28.self_attn.k_proj
- model.layers.36.self_attn.k_proj
- model.layers.25.self_attn.k_proj
- model.layers.37.self_attn.k_proj
- model.layers.30.self_attn.k_proj
- model.layers.16.self_attn.k_proj
- model.layers.32.self_attn.k_proj
- model.layers.41.self_attn.k_proj
- model.layers.26.self_attn.k_proj
# self_attn.o_proj layers
- model.layers.50.self_attn.o_proj
- model.layers.61.self_attn.o_proj
- model.layers.46.self_attn.o_proj
- model.layers.53.self_attn.o_proj
- model.layers.54.self_attn.o_proj
- model.layers.19.self_attn.o_proj
- model.layers.42.self_attn.o_proj
- model.layers.49.self_attn.o_proj
- model.layers.41.self_attn.o_proj
- model.layers.68.self_attn.o_proj
- model.layers.18.self_attn.o_proj
- model.layers.45.self_attn.o_proj
- model.layers.11.self_attn.o_proj
- model.layers.67.self_attn.o_proj
- model.layers.48.self_attn.o_proj
- model.layers.51.self_attn.o_proj
- model.layers.64.self_attn.o_proj
- model.layers.13.self_attn.o_proj
- model.layers.14.self_attn.o_proj
- model.layers.16.self_attn.o_proj
- model.layers.17.self_attn.o_proj
- model.layers.47.self_attn.o_proj
- model.layers.0.self_attn.o_proj
- model.layers.20.self_attn.o_proj
- model.layers.63.self_attn.o_proj
- model.layers.15.self_attn.o_proj
- model.layers.5.self_attn.o_proj
- model.layers.21.self_attn.o_proj
- model.layers.52.self_attn.o_proj
- model.layers.12.self_attn.o_proj
- model.layers.10.self_attn.o_proj
- model.layers.62.self_attn.o_proj
- model.layers.56.self_attn.o_proj
- model.layers.22.self_attn.o_proj
- model.layers.6.self_attn.o_proj
- model.layers.7.self_attn.o_proj
# self_attn.q_proj layers
- model.layers.2.self_attn.q_proj
- model.layers.4.self_attn.q_proj
- model.layers.46.self_attn.q_proj
- model.layers.5.self_attn.q_proj
- model.layers.7.self_attn.q_proj
- model.layers.6.self_attn.q_proj
- model.layers.9.self_attn.q_proj
- model.layers.10.self_attn.q_proj
- model.layers.1.self_attn.q_proj
- model.layers.18.self_attn.q_proj
- model.layers.62.self_attn.q_proj
- model.layers.8.self_attn.q_proj
- model.layers.15.self_attn.q_proj
- model.layers.14.self_attn.q_proj
- model.layers.16.self_attn.q_proj
- model.layers.31.self_attn.q_proj
- model.layers.19.self_attn.q_proj
- model.layers.17.self_attn.q_proj
- model.layers.33.self_attn.q_proj
- model.layers.35.self_attn.q_proj
- model.layers.12.self_attn.q_proj
- model.layers.21.self_attn.q_proj
- model.layers.27.self_attn.q_proj
- model.layers.34.self_attn.q_proj
- model.layers.13.self_attn.q_proj
- model.layers.56.self_attn.q_proj
- model.layers.11.self_attn.q_proj
- model.layers.52.self_attn.q_proj
- model.layers.54.self_attn.q_proj
- model.layers.28.self_attn.q_proj
- model.layers.30.self_attn.q_proj
- model.layers.20.self_attn.q_proj
- model.layers.29.self_attn.q_proj
- model.layers.37.self_attn.q_proj
- model.layers.23.self_attn.q_proj
- model.layers.75.self_attn.q_proj
# self_attn.v_proj layers
- model.layers.11.self_attn.v_proj
- model.layers.17.self_attn.v_proj
- model.layers.37.self_attn.v_proj
- model.layers.40.self_attn.v_proj
- model.layers.41.self_attn.v_proj
- model.layers.42.self_attn.v_proj
- model.layers.43.self_attn.v_proj
- model.layers.44.self_attn.v_proj
- model.layers.45.self_attn.v_proj
- model.layers.46.self_attn.v_proj
- model.layers.48.self_attn.v_proj
- model.layers.49.self_attn.v_proj
- model.layers.50.self_attn.v_proj
- model.layers.51.self_attn.v_proj
- model.layers.53.self_attn.v_proj
- model.layers.54.self_attn.v_proj
- model.layers.55.self_attn.v_proj
- model.layers.57.self_attn.v_proj
- model.layers.58.self_attn.v_proj
- model.layers.59.self_attn.v_proj
- model.layers.60.self_attn.v_proj
- model.layers.61.self_attn.v_proj
- model.layers.62.self_attn.v_proj
- model.layers.63.self_attn.v_proj
- model.layers.64.self_attn.v_proj
- model.layers.65.self_attn.v_proj
- model.layers.66.self_attn.v_proj
- model.layers.67.self_attn.v_proj
- model.layers.69.self_attn.v_proj
- model.layers.75.self_attn.v_proj
- model.layers.18.self_attn.v_proj
- model.layers.78.self_attn.v_proj
- model.layers.68.self_attn.v_proj
- model.layers.47.self_attn.v_proj
- model.layers.38.self_attn.v_proj
- model.layers.71.self_attn.v_proj
# model.norm layers



dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: /workspace/axolotl/llama-70b

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

wandb_project: llama-3
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 3
optimizer: adamw_8bit
lr_scheduler: cosine
learning_rate: 1e-5

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

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 5
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 4
save_total_limit: 2
save_steps:
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
weight_decay: 0.00
fsdp:
fsdp_config:
special_tokens:
  eos_token: "<|im_end|>"
  pad_token: "<|end_of_text|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"

</details><br>

workspace/axolotl/llama-70b

This model is a fine-tuned version of meta-llama/Meta-Llama-3-70B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4808

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: 1e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 5
• num_epochs: 3

Training results

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.2+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1