70B-L3.3-mhnnn-x1-FP8-Dynamic

This quant was made for and by Infermatic.ai

Sao10K/70B-L3.3-mhnnn-x1

Copy of the original card

---

my mental when things do not go well

70B-L3.3-mhnnn-x1

I quite liked it, after messing around. Same data composition as Freya, applied differently.

Has occasional brainfarts which are fixed with a regen, the price for more creative outputs.

Recommended Model Settings | Look, I just use these, they work fine enough. I don't even know how DRY or other meme samplers work. Your system prompt matters more anyway.

Prompt Format: Llama-3-Instruct
Temperature: 1.1
min_p: 0.05

Types of Data included within Sets

Completion - Novels / eBooks
Text Adventure - Include details like 'Text Adventure Narrator' in the System Prompt, give it a one-shot example and it'll fly.
Amoral Assistant - Include the terms 'Amoral', 'Neutral' along with the regular assistant prompt for better results.
Instruct / Assistant - The usual assistant tasks.
Roleplay - As per Usual, Regular Sets 

Training time in total was ~14 Hours on a 8xH100 Node, shout out to SCDF for not sponsoring this run. My funds are dry doing random things.

https://sao10k.carrd.co/ for contact.

---

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

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

axolotl version: 0.6.0

adapter: lora # 16-bit
lora_r: 64
lora_alpha: 64
lora_dropout: 0.2
peft_use_rslora: true
lora_target_linear: true
  
# Data
dataset_prepared_path: dataset_run_freya
datasets:
# S1 - Writing / Completion
  - path: datasets/eBooks-cleaned-75K
    type: completion
  - path: datasets/novels-clean-dedupe-10K
    type: completion
# S2 - Instruct
  - path: datasets/10k-amoral-full-fixed-sys.json
    type: chat_template
    chat_template: llama3
    roles_to_train: ["gpt"]
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    train_on_eos: turn
  - path: datasets/44k-hespera-smartshuffle.json
    type: chat_template
    chat_template: llama3
    roles_to_train: ["gpt"]
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    train_on_eos: turn
  - path: datasets/5k_rpg_adventure_instruct-sys.json
    type: chat_template
    chat_template: llama3
    roles_to_train: ["gpt"]
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    train_on_eos: turn
shuffle_merged_datasets: true
warmup_ratio: 0.1

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: true

# Iterations
num_epochs: 1

# Sampling
sample_packing: true
pad_to_sequence_len: true
train_on_inputs: false
group_by_length: false

# Batching
gradient_accumulation_steps: 4
micro_batch_size: 2
gradient_checkpointing: unsloth

# Evaluation
val_set_size: 0.025
evals_per_epoch: 5
eval_table_size:
eval_max_new_tokens: 256
eval_sample_packing: false
eval_batch_size: 1

# Optimizer
optimizer: paged_ademamix_8bit
lr_scheduler: cosine
learning_rate: 0.00000242
weight_decay: 0.2
max_grad_norm: 10.0

# Garbage Collection
gc_steps: 10

# Misc
deepspeed: ./deepspeed_configs/zero3_bf16.json

</details><br>