See axolotl config

axolotl version: 0.4.0

base_model: mistralai/Mistral-7B-v0.1
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizergin
is_mistral_derived_model: true

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  # This will be the path used for the data when it is saved to the Volume in the cloud.
  - path: data.jsonl
    ds_type: json
    type:
      # JSONL file contains question, context, answer fields per line.
      # This gets mapped to instruction, input, output axolotl tags.
      field_instruction: instruction
      field_input: input
      field_output: output
      # Format is used by axolotl to generate the prompt.
      format: |-
        [INST]{input}
        {instruction} [/INST] 

dataset_prepared_path:
val_set_size: 0.05
output_dir: ./lora-out

sequence_len: 4096
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: false

adapter: lora
lora_model_dir:
lora_r: 16
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_run_id:

gradient_accumulation_steps: 1
micro_batch_size: 32
num_epochs: 4
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.0001

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

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

warmup_steps: 10
save_steps:
debug:
deepspeed: /root/axolotl/deepspeed_configs/zero3_bf16.json
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

Mistral Sentiment Analysis

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the FinGPT Sentiment dataset. It is intended to be used for sentiment analysis tasks for financial data. Data was modified to use with Axolotl, see here for the modified data. See the FinGPT Project for more information. It achieves the following results on the evaluation set:

• Loss: 0.1598

Ollama Example

ollama run chand1012/mistral_sentiment
>>> Apple (NASDAQ:AAPL) Up Fractionally despite Rising Vision Pro Returns Please choose an answer from {negative/neutral/positive} 
 positive

Python Example

from transformers import AutoModel, AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM, LlamaTokenizerFast
from peft import PeftModel  # 0.8.2

# Load Models
base_model = "mistralai/Mistral-7B-v0.1" 
peft_model = "TimeSurgeLabs/mistral_sentiment_lora"
tokenizer = LlamaTokenizerFast.from_pretrained(base_model, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
model = LlamaForCausalLM.from_pretrained(base_model, trust_remote_code=True, device_map = "cuda:0", load_in_8bit = True,)
model = PeftModel.from_pretrained(model, peft_model)
model = model.eval()

# Make prompts
prompt = [
'''Instruction: What is the sentiment of this news? Please choose an answer from {negative/neutral/positive}
Input: FINANCING OF ASPOCOMP 'S GROWTH Aspocomp is aggressively pursuing its growth strategy by increasingly focusing on technologically more demanding HDI printed circuit boards PCBs .
Answer: ''',
'''Instruction: What is the sentiment of this news? Please choose an answer from {negative/neutral/positive}
Input: According to Gran , the company has no plans to move all production to Russia , although that is where the company is growing .
Answer: ''',
'''Instruction: What is the sentiment of this news? Please choose an answer from {negative/neutral/positive}
Input: A tinyurl link takes users to a scamming site promising that users can earn thousands of dollars by becoming a Google ( NASDAQ : GOOG ) Cash advertiser .
Answer: ''',
]

# Generate results
tokens = tokenizer(prompt, return_tensors='pt', padding=True, max_length=512)
res = model.generate(**tokens, max_length=512)
res_sentences = [tokenizer.decode(i) for i in res]
out_text = [o.split("Answer: ")[1] for o in res_sentences]

# show results
for sentiment in out_text:
    print(sentiment)

# Output:    
# positive
# neutral
# negative

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 64
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9, 0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 4

Training results

Framework versions

• PEFT 0.8.2
• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.17.0
• Tokenizers 0.15.0