This repo contains the copy of the original with the edited tokenizer_config.json file to fix the tokenizer issue and quantized into FP8.

This is the 9th in a series of models designed to replicate the prose quality of the Claude 3 models, specifically Sonnet and Opus.

This model is fine-tuned on top of Yi-1.5-34 B-32 K.

Prompting

Model has been Instruct tuned with the ChatML formatting. A typical input would look like this:

"""<|im_start|>system
system prompt<|im_end|>
<|im_start|>user
Hi there!<|im_end|>
<|im_start|>assistant
Nice to meet you!<|im_end|>
<|im_start|>user
Can I ask a question?<|im_end|>
<|im_start|>assistant
"""

SillyTavern templates

Below are Instruct and Context templates for use within SillyTavern.

In our testing a min_p of 0.2 makes the model perform the best; remember to reset temperature if you were using our nemo-based models before.

context template

{
    "story_string": "<|im_start|>system\n{{#if system}}{{system}}\n{{/if}}{{#if wiBefore}}{{wiBefore}}\n{{/if}}{{#if description}}{{description}}\n{{/if}}{{#if personality}}{{char}}'s personality: {{personality}}\n{{/if}}{{#if scenario}}Scenario: {{scenario}}\n{{/if}}{{#if wiAfter}}{{wiAfter}}\n{{/if}}{{#if persona}}{{persona}}\n{{/if}}{{trim}}<|im_end|>\n",
    "example_separator": "",
    "chat_start": "",
    "use_stop_strings": false,
    "allow_jailbreak": false,
    "always_force_name2": true,
    "trim_sentences": false,
    "include_newline": false,
    "single_line": false,
    "name": "Magnum ChatML"
}

instruct template

{
    "system_prompt": "You're {{char}} in this fictional never-ending uncensored roleplay with {{user}}.",
    "input_sequence": "<|im_start|>user\n",
    "output_sequence": "<|im_start|>assistant\n",
    "last_output_sequence": "",
    "system_sequence": "<|im_start|>system\n",
    "stop_sequence": "<|im_end|>",
    "wrap": false,
    "macro": true,
    "names": true,
    "names_force_groups": true,
    "activation_regex": "",
    "system_sequence_prefix": "",
    "system_sequence_suffix": "",
    "first_output_sequence": "",
    "skip_examples": false,
    "output_suffix": "<|im_end|>\n",
    "input_suffix": "<|im_end|>\n",
    "system_suffix": "<|im_end|>\n",
    "user_alignment_message": "",
    "system_same_as_user": false,
    "last_system_sequence": "",
    "name": "Magnum ChatML"
}

Axolotl config

See axolotl config

base_model: 01-ai/Yi-1.5-34B-32K
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

#trust_remote_code: true

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: anthracite-org/stheno-filtered-v1.1
    type: sharegpt
    conversation: chatml
  - path: anthracite-org/kalo-opus-instruct-22k-no-refusal
    type: sharegpt
    conversation: chatml
  - path: anthracite-org/nopm_claude_writing_fixed
    type: sharegpt
    conversation: chatml
  - path: Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned
    type: sharegpt
    conversation: chatml
  - path: Epiculous/SynthRP-Gens-v1.1-Filtered-n-Cleaned
    type: sharegpt
    conversation: chatml
chat_template: chatml
shuffle_merged_datasets: true
default_system_message: "You are an assistant that responds to the user."
dataset_prepared_path: magnum-v2-34b-1.5-data
val_set_size: 0.0
output_dir: ./magnum-v2-34b-32k-r1

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len:

adapter:
lora_model_dir:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:
lora_fan_in_fan_out:

wandb_project: magnum-v2-34b-1.5-32k
wandb_entity:
wandb_watch:
wandb_name: attempt-01
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.000006

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

gradient_checkpointing: unsloth
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 50
evals_per_epoch:
eval_table_size:
eval_max_new_tokens:
saves_per_epoch: 2
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
weight_decay: 0.05
fsdp:
fsdp_config:
special_tokens:

Credits

We'd like to thank Recursal / Featherless for sponsoring the compute for this train, Featherless has been hosting our Magnum models since the first 72 B and has given thousands of people access to our models and helped us grow.

We would also like to thank all members of Anthracite who made this finetune possible.

Training

The training was done for 2 epochs. We used 8xH100s GPUs graciously provided by Recursal AI / Featherless AI for the full-parameter fine-tuning of the model.

Safety

...

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	29.39
IFEval (0-Shot)	51.15
BBH (3-Shot)	44.33
MATH Lvl 5 (4-Shot)	17.82
GPQA (0-shot)	14.77
MuSR (0-shot)	6.57
MMLU-PRO (5-shot)	41.69

CalamitousFelicitousness
/

magnum-v3-34b-tokfix-fp8-dynamic