olympiadmind-gemma-physics

QLoRA adapter (rank 16, alpha 32) on google/gemma-2-2b-it that lists the physics concepts needed to solve an olympiad problem. Part of OlympiadMind.

Trainable params 6.4M / 2.6B (0.24%)
Training loss 2.94 β†’ 1.30
Token accuracy 48% β†’ 70%
Hardware RTX 3050 6GB, 41 min
Data olympiadmind-data concept labels

Use

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

adapter = "amanDS5153/olympiadmind-gemma-physics"
tok   = AutoTokenizer.from_pretrained(adapter)
base  = AutoModelForCausalLM.from_pretrained("google/gemma-2-2b-it", dtype=torch.float16, device_map="cuda")
model = PeftModel.from_pretrained(base, adapter)

problem = "A charged particle enters a uniform magnetic field B with velocity v. Find the radius of its path."
prompt  = f"<start_of_turn>user\nList the physics concepts required to solve this problem.\n\n{problem}<end_of_turn>\n<start_of_turn>model\n"
ids = tok(prompt, return_tensors="pt").to("cuda")
out = model.generate(**ids, max_new_tokens=100, do_sample=False)
print(tok.decode(out[0][ids["input_ids"].shape[1]:], skip_special_tokens=True))

Gemma is provided under the Gemma Terms of Use.

Downloads last month
21
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for amanDS5153/olympiadmind-gemma-physics

Adapter
(541)
this model