---
tags:
- physics
- cosmology
model-index:
- name: cosmosage_qa
results: []
license: mit
language:
- en
pipeline_tag: text-generation
base_model: mistralai/Mistral-7B-v0.1
---
# cosmosage
Cosmosage is a natural-language cosmology assistant that can answer questions about cosmology.
cosmosage_v2 first underwent continued pretraining based on thousands of papers and textbooks,
and was subsequently fine-tuned on synthetically-generated question-answer pairs. It is a full
chat model, though it excels in Q&A mode, where the model gives a single answer in response to
a single question.
The code used to generate cosmosage_v2 is available at https://github.com/tijmen/cosmosage
## Usage
After downloading cosmosage_v2, the following example code can be used to ask questions:
```python
path_to_model = 'cosmosage_v2/'
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
device = "cuda"
model = AutoModelForCausalLM.from_pretrained(path_to_model).to(device)
tokenizer = AutoTokenizer.from_pretrained(path_to_model)
def ask_cosmosage(question):
input_ids = torch.cat([
tokenizer.encode("You are cosmosage, an AI programmed to be a cosmology expert. You answer the USER's question clearly in long form, always providing context. When appropriate, provide a reference.", return_tensors="pt"),
torch.tensor([[28705]]),
tokenizer.encode("USER:", add_special_tokens=False, return_tensors="pt"),
tokenizer.encode(question, add_special_tokens=False, return_tensors="pt"),
torch.tensor([[28705]]),
tokenizer.encode("ASSISTANT:", add_special_tokens=False, return_tensors="pt")
], dim=-1).to(device)
generated_ids = model.generate(input_ids, max_length=input_ids.shape[1] + 1000, do_sample=True)
return tokenizer.decode(generated_ids[0], skip_special_tokens=True)```
## Comparison to cosmosage_v1
cosmosage_v2 is a more knowledgeable model than cosmosage_v1 due to being pretrained on the papers and
textbooks, rather than just on synthetically generated QA pairs. However, it continues to struggle with
_reliability_. While many of its answers are factually accurate, some are not. The outputs of cosmosage
(or any LLM) should not be trusted to be factual.
[
](https://github.com/OpenAccess-AI-Collective/axolotl)
See axolotl config
axolotl version: `0.4.0`
```yaml
base_model: /workspace/output/cosmosage_base/
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true
load_in_8bit: false
load_in_4bit: false
strict: false
datasets:
- path: /workspace/input/datasets/qa_tune/arxiv_metadata_qa3.jsonl
type: sharegpt
- path: /workspace/input/datasets/qa_tune/arxiv_refined_qa.jsonl
type: sharegpt
- path: /workspace/input/datasets/qa_tune/arxiv_summary3.jsonl
type: sharegpt
- path: /workspace/input/datasets/qa_tune/cosmology_qa.jsonl
type: alpaca_chat.load_qa
- path: /workspace/input/datasets/qa_tune/openhermes2_5.jsonl
type: sharegpt
- path: /workspace/input/datasets/qa_tune/cosmology_textbooks_qa.jsonl
type: alpaca_chat.load_qa
- path: /workspace/input/datasets/qa_tune/physics_astro_qa.jsonl
type: alpaca_chat.load_qa
dataset_prepared_path: /workspace/output/qa_tune_prepared
val_set_size: 0.001
output_dir: /workspace/output/cosmosage_qa
chat_template: inst
adapter:
lora_model_dir:
sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true
lora_r:
lora_alpha:
lora_dropout:
lora_target_modules:
lora_target_linear:
lora_fan_in_fan_out:
seed: 702
wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
gradient_accumulation_steps: 1
micro_batch_size: 4
num_epochs: 2.0
optimizer: adamw_torch
lr_scheduler: linear
learning_rate: 0.000002
max_grad_norm: 3.0
train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false
gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
warmup_steps: 100
eval_steps: 0.05
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
save_total_limit: 2
debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero1.json
weight_decay:
fsdp:
fsdp_config:
special_tokens:
bos_token: ""
eos_token: ""
unk_token: ""
ddp_timeout: 7200000
```
# workspace/output/cosmosage_qa
This model was trained from scratch on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5673
## 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-06
- train_batch_size: 4
- eval_batch_size: 4
- seed: 702
- distributed_type: multi-GPU
- num_devices: 4
- total_train_batch_size: 16
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 100
- num_epochs: 2.0
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:-----:|:---------------:|
| 1.1004 | 0.0 | 1 | 1.1450 |
| 0.7343 | 0.1 | 909 | 0.7093 |
| 0.697 | 0.2 | 1818 | 0.6630 |
| 0.6386 | 0.3 | 2727 | 0.6380 |
| 0.5687 | 0.4 | 3636 | 0.6212 |
| 0.5857 | 0.5 | 4545 | 0.6083 |
| 0.6161 | 0.6 | 5454 | 0.5986 |
| 0.522 | 0.7 | 6363 | 0.5894 |
| 0.5563 | 0.8 | 7272 | 0.5825 |
| 0.6176 | 0.9 | 8181 | 0.5766 |
| 0.5948 | 1.0 | 9090 | 0.5719 |
| 0.4269 | 1.08 | 9999 | 0.5817 |
| 0.4858 | 1.18 | 10908 | 0.5796 |
| 0.4909 | 1.28 | 11817 | 0.5765 |
| 0.4325 | 1.38 | 12726 | 0.5746 |
| 0.4037 | 1.48 | 13635 | 0.5720 |
| 0.507 | 1.58 | 14544 | 0.5706 |
| 0.4778 | 1.68 | 15453 | 0.5697 |
| 0.4599 | 1.78 | 16362 | 0.5683 |
| 0.4515 | 1.88 | 17271 | 0.5673 |
### Framework versions
- Transformers 4.38.0.dev0
- Pytorch 2.0.1+cu118
- Datasets 2.17.0
- Tokenizers 0.15.0