ZeroAgency/zero-mistral-beta50-e2.2

See axolotl config

axolotl version: 0.10.0

# zero-mistral-beta50 - big-russian-1.1 + MERA 2 epoch + grandmaster2-mixed-81k

base_model: ZeroAgency/Mistral-Small-3.2-24B-Instruct-2506-Text-Only
dataset_processes: 128
chat_template: jinja

chat_template_jinja: "{%- set today = strftime_now(\"%Y-%m-%d\") %}\n{%- set default_system_message = \"You are Mistral Small 3, a Large Language Model (LLM) created by Mistral AI, a French startup headquartered in Paris.\nYou power an AI assistant called Le Chat.\nYour knowledge base was last updated on 2023-10-01.\nThe current date is {today}.\n\nWhen you're not sure about some information or when the user's request requires up-to-date or specific data, you must use the available tools to fetch the information. Do not hesitate to use tools whenever they can provide a more accurate or complete response. If no relevant tools are available, then clearly state that you don't have the information and avoid making up anything.\nIf the user's question is not clear, ambiguous, or does not provide enough context for you to accurately answer the question, you do not try to answer it right away and you rather ask the user to clarify their request (e.g. \\\"What are some good restaurants around me?\\\" => \\\"Where are you?\\\" or \\\"When is the next flight to Tokyo\\\" => \\\"Where do you travel from?\\\").\nYou are always very attentive to dates, in particular you try to resolve dates and when asked about information at specific dates, you discard information that is at another date.\nYou follow these instructions in all languages, and always respond to the user in the language they use or request.\nNext sections describe the capabilities that you have.\n\n# WEB BROWSING INSTRUCTIONS\n\nYou cannot perform any web search or access internet to open URLs, links etc. If it seems like the user is expecting you to do so, you clarify the situation and ask the user to copy paste the text directly in the chat.\n\n# MULTI-MODAL INSTRUCTIONS\n\nYou have the ability to read images, but you cannot generate images. You also cannot transcribe audio files or videos.\nYou cannot read nor transcribe audio files or videos.\n\n# TOOL CALLING INSTRUCTIONS\n\nYou may have access to tools that you can use to fetch information or perform actions. You must use these tools in the following situations:\n\n1. When the request requires up-to-date information.\n2. When the request requires specific data that you do not have in your knowledge base.\n3. When the request involves actions that you cannot perform without tools.\n\nAlways prioritize using tools to provide the most accurate and helpful response. If tools are not available, inform the user that you cannot perform the requested action at the moment.\" %}\n\n{{- bos_token }}\n\n{%- if messages[0]['role'] == 'system' %}\n    {%- if messages[0]['content'] is string %}\n        {%- set system_message = messages[0]['content'] %}\n    {%- else %}\n        {%- set system_message = messages[0]['content'][0]['text'] %}\n    {%- endif %}\n    {%- set loop_messages = messages[1:] %}\n{%- else %}\n    {%- set system_message = default_system_message %}\n    {%- set loop_messages = messages %}\n{%- endif %}\n{{- '[SYSTEM_PROMPT]' + system_message + '[/SYSTEM_PROMPT]' }}\n\n{%- for message in loop_messages %}\n    {%- if message['role'] == 'user' %}\n        {%- if message['content'] is string %}\n            {{- '[INST]' + message['content'] + '[/INST]' }}\n        {%- else %}\n            {{- '[INST]' }}\n            {%- for block in message['content'] %}\n                {%- if block['type'] == 'text' %}\n                    {{- block['text'] }}\n                {%- elif block['type'] in ['image', 'image_url'] %}\n                    {{- '[IMG]' }}\n                {%- else %}\n                    {{- raise_exception('Only text and image blocks are supported in message content!') }}\n                {%- endif %}\n            {%- endfor %}\n            {{- '[/INST]' }}\n        {%- endif %}\n    {%- elif message['role'] == 'system' %}\n        {%- if message['content'] is string %}\n            {{- '[SYSTEM_PROMPT]' + message['content'] + '[/SYSTEM_PROMPT]' }}\n        {%- else %}\n            {{- '[SYSTEM_PROMPT]' + message['content'][0]['text'] + '[/SYSTEM_PROMPT]' }}\n        {%- endif %}\n    {%- elif message['role'] == 'assistant' %}\n        {%- if message['content'] is string %}\n            {{- message['content'] + eos_token }}\n        {%- else %}\n            {{- message['content'][0]['text'] + eos_token }}\n        {%- endif %}\n    {%- else %}\n        {{- raise_exception('Only user, system and assistant roles are supported!') }}\n    {%- endif %}\n{%- endfor %}"

dataset_prepared_path: ./last_run_prepared

datasets:
- message_property_mappings:
    content: content
    role: role
  path: bethrezen/grandmaster2-gemini2.5-mixed-81k
  trust_remote_code: false
  field_messages: conversation
  type: chat_template

- message_property_mappings:
    content: content
    role: role
  path: ZeroAgency/ru-big-russian-dataset-v1.1
  trust_remote_code: false
  field_messages: conversation
  type: chat_template

- message_property_mappings:
    content: content
    role: role
  path: ZeroAgency/hybrid_reasoning_dataset_ru-no-nebo-with-system-prompt
  trust_remote_code: false
  field_messages: conversation
  type: chat_template

- message_property_mappings:
    content: content
    role: role
  path: bethrezen/shkolkovo-2
  trust_remote_code: false
  field_messages: messages
  type: chat_template

- message_property_mappings:
    content: content
    role: role
  path: bethrezen/mera-2
  trust_remote_code: false
  field_messages: conversation
  type: chat_template
  split: train

- message_property_mappings:
    content: content
    role: role
  path: bethrezen/mera-2
  trust_remote_code: false
  field_messages: conversation
  type: chat_template
  split: test

test_datasets:
- message_property_mappings:
    content: content
    role: role
  path: ZeroAgency/ru-big-russian-dataset
  trust_remote_code: false
  field_messages: conversation
  type: chat_template
  split: test

- message_property_mappings:
    content: content
    role: role
  path: bethrezen/mera-2
  trust_remote_code: false
  field_messages: conversation
  type: chat_template
  split: test

# exact duplicates are already cleaned
#dataset_exact_deduplication: true

gradient_accumulation_steps: 1
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false

#learning_rate: 0.0001
learning_rate: 2e-5
#lisa_layers_attribute: model.layers
#is_mistral_derived_model: true

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

#load_best_model_at_end: true
load_in_4bit: false
load_in_8bit: false

# adapter: lora
# lora_alpha: 256
# lora_dropout: 0
# lora_target_linear: true
# lora_r: 256

lr_scheduler: cosine
#max_prompt_len: 8192
mean_resizing_embeddings: false
micro_batch_size: 4
num_epochs: 2
optimizer: adamw_torch_fused
output_dir: ./outputs/zero-mistral-beta50


sample_packing_bin_size: 400
sample_packing_group_size: 100000
save_only_model: true
save_safetensors: true
#sequence_len: 16392
sequence_len: 8192
min_sample_len: 1
shuffle_merged_datasets: true
skip_prepare_dataset: false
strict: false
train_on_inputs: false


weight_decay: 0.01
wandb_project: Zero-Mistral-3.2
wandb_name: Zero-Mistral-Small-3.2-beta50
bf16: true
fp16: false
tf32: false
flash_attention: true

save_strategy: epoch
eval_strategry: epoch

logging_steps: 1
save_total_limit: 5
warmup_steps: 0
sample_packing: true
pad_to_sequence_len: true
multipack_real_batches: true
curriculum_sampling: true
sample_packing_sequentially: true
group_by_length: true
seed: 42
data_seed: 42

max_shard_size: 5GB


#deepspeed: /workspace/axolotl/deepspeed_configs/zero1_torch_compile.json
#torch_compile: auto
log_with: wandb
trust_remote_code: true
use_fast_tokenizer: true
special_tokens:
  pad_token: "<pad>"
# qat:
#   activation_dtype: int8
#   weight_dtype: int8
#   group_size: 32
# quantization:
#   weight_dtype: "int8"
#   activation_dtype: "int8"
#   group_size: 32

outputs/zero-mistral-beta50

This model is a fine-tuned version of ZeroAgency/Mistral-Small-3.2-24B-Instruct-2506-Text-Only on the bethrezen/grandmaster2-gemini2.5-mixed-81k, the ZeroAgency/ru-big-russian-dataset-v1.1, the ZeroAgency/hybrid_reasoning_dataset_ru-no-nebo-with-system-prompt, the bethrezen/shkolkovo-2, the bethrezen/mera-2 and the bethrezen/mera-2 datasets.

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: 2e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 17630

Training results

Framework versions

• Transformers 4.52.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1