Instructions to use cs-552-2026-MandMP/math_model with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cs-552-2026-MandMP/math_model with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="cs-552-2026-MandMP/math_model")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cs-552-2026-MandMP/math_model")
model = AutoModelForCausalLM.from_pretrained("cs-552-2026-MandMP/math_model", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use cs-552-2026-MandMP/math_model with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cs-552-2026-MandMP/math_model"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cs-552-2026-MandMP/math_model",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cs-552-2026-MandMP/math_model

SGLang

How to use cs-552-2026-MandMP/math_model with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "cs-552-2026-MandMP/math_model" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cs-552-2026-MandMP/math_model",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "cs-552-2026-MandMP/math_model" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cs-552-2026-MandMP/math_model",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cs-552-2026-MandMP/math_model with Docker Model Runner:
```
docker model run hf.co/cs-552-2026-MandMP/math_model
```

math_model

Commit History

Jun 3

09:24
Automated MNLP evaluation report (2026-06-03) (#14)⁠

sohaddad +1: zechen-nlp

a41c3c3

Jun 1

11:36
Automated MNLP evaluation report (2026-06-01) (#13)⁠

sohaddad +1: zechen-nlp

0955d60

May 29

09:22
Automated MNLP evaluation report (2026-05-29) (#12)⁠

sohaddad +1: zechen-nlp

58b649f

May 28

21:17
concise_sft_v1: SFT on 7899 concise traces from Qwen3-8B (thinking off), 1 epoch on top of dapo_v3

sohaddad

Verified signature
051ffca
08:55
Automated MNLP evaluation report (2026-05-28) (#11)⁠

sohaddad +1: zechen-nlp

f58c2f0

May 27

14:44
Automated MNLP evaluation report (2026-05-27) (#10)⁠

sohaddad +1: zechen-nlp

2c97fe4

May 26

18:50
Automated MNLP evaluation report (2026-05-26) (#9)⁠

sohaddad +1: zechen-nlp

e95e6f9

May 25

22:49
Upload dapo_v3_merged_latest weights

sohaddad

Verified signature
46b0423
10:13
Add repetition_penalty=1.05

sohaddad

Verified signature
c1721de
10:07
Automated MNLP evaluation report (2026-05-25) (#8)⁠

sohaddad +1: zechen-nlp

d660f80

May 24

20:52
Automated MNLP evaluation report (2026-05-24) (#7)⁠

sohaddad +1: zechen-nlp

068d0e5
00:40
DAPO RL 1600 steps merged

sohaddad

Verified signature
1418c00
00:35
Automated MNLP evaluation report (2026-05-23) (#6)⁠

sohaddad +1: zechen-nlp

5a568eb

May 22

09:06
DAPO RL 800 steps merged

sohaddad

Verified signature
5a9ca7a

May 21

20:23
Upload dapo_v2 merged checkpoint at step 200

sohaddad

Verified signature
88c1571
08:27
Automated MNLP evaluation report (2026-05-21) (#5)⁠

sohaddad +1: zechen-nlp

335c35a

May 20

14:54
Add checkpoint-580: SFT distill with math system prompt + enable_thinking

sohaddad

Verified signature
6e0bf51
08:03
Fix tokenizer_config: set pad_token to <|im_end|>

sohaddad

Verified signature
796af15
08:03
Fix chat_template: add enable_thinking=true and math system prompt auto-injection

sohaddad

Verified signature
7b6d984
07:47
Automated MNLP evaluation report (2026-05-20) (#4)⁠

sohaddad +1: zechen-nlp

ba906aa

May 19

11:42
fix: default enable_thinking to ON by adding else branch for <think> injection

sohaddad

Verified signature
0f88a28
11:32
fix: update generation_config.json to match project spec

sohaddad

Verified signature
264aa12
11:04
fix: set pad_token_id to 151643 (endoftext)

sohaddad

Verified signature
46c3f56
08:23
Fix: restore Qwen3 chat template

sohaddad

Verified signature
95d0da9
08:23
Fix: restore Qwen3 chat template

sohaddad

Verified signature
bcc275e
08:03
Automated MNLP evaluation report (2026-05-19) (#3)⁠

sohaddad +1: zechen-nlp

dc38dbd

May 18

11:48
Fix generation_config: correct pad_token_id, add bos_token_id, remove training-time overrides

sohaddad

Verified signature
3a4ce59
11:03
Automated MNLP evaluation report (2026-05-18) (#2)⁠

sohaddad +1: zechen-nlp

63c4966

May 17

21:35
Remove training_args.bin (training artifact, not needed for inference)

sohaddad

Verified signature
b455b43
21:35
Remove trainer_state.json (training artifact, not needed for inference)

sohaddad

Verified signature
eccf3f4
21:35
Remove scheduler.pt (training artifact, not needed for inference)

sohaddad

Verified signature
0b52c5b
21:35
Remove rng_state.pth (training artifact, not needed for inference)

sohaddad

Verified signature
9584bd0
21:35
Remove optimizer.pt (training artifact, not needed for inference)

sohaddad

Verified signature
0776556
21:33
Distilled checkpoint (Qwen3-8B teacher, MATH train, 2 epochs)

sohaddad

Verified signature
6f45883

May 13

13:06
Automated MNLP evaluation report (2026-05-13) (#1)⁠

sohaddad +1: zechen-nlp

b90a696

May 12

20:05
Add thinking_token_budget=1500 to generation config

sohaddad

Verified signature
bec4dc9
19:57
Fix system prompt with correct boxed format

sohaddad

Verified signature
650e0ad
19:56
Add default system prompt for boxed format compliance

sohaddad

Verified signature
c79a414
19:45
Remove training_args.bin artifact

sohaddad

Verified signature
d04b411
19:44
Restore max_new_tokens in generation_config

sohaddad

Verified signature
fdeb43f
19:44
Remove training artifacts and checkpoint folders

sohaddad

Verified signature
5299acc
19:40
Upload folder using huggingface_hub

sohaddad

Verified signature
bf7cb5b
15:31
Upload tokenizer

sohaddad

Verified signature
a141847
15:31
Upload Qwen3ForCausalLM

sohaddad

Verified signature
609e9b1
14:35
initial commit

sohaddad

Verified signature
56002cc

Commit History

Jun 3

Automated MNLP evaluation report (2026-06-03) (#14)⁠

Jun 1

Automated MNLP evaluation report (2026-06-01) (#13)⁠

May 29

Automated MNLP evaluation report (2026-05-29) (#12)⁠

May 28

concise_sft_v1: SFT on 7899 concise traces from Qwen3-8B (thinking off), 1 epoch on top of dapo_v3

Automated MNLP evaluation report (2026-05-28) (#11)⁠

May 27

Automated MNLP evaluation report (2026-05-27) (#10)⁠

May 26

Automated MNLP evaluation report (2026-05-26) (#9)⁠

May 25

Upload dapo_v3_merged_latest weights

Add repetition_penalty=1.05

Automated MNLP evaluation report (2026-05-25) (#8)⁠

May 24

Automated MNLP evaluation report (2026-05-24) (#7)⁠

DAPO RL 1600 steps merged

Automated MNLP evaluation report (2026-05-23) (#6)⁠

May 22

DAPO RL 800 steps merged

May 21

Upload dapo_v2 merged checkpoint at step 200

Automated MNLP evaluation report (2026-05-21) (#5)⁠

May 20

Add checkpoint-580: SFT distill with math system prompt + enable_thinking

Fix tokenizer_config: set pad_token to <|im_end|>

Fix chat_template: add enable_thinking=true and math system prompt auto-injection

Automated MNLP evaluation report (2026-05-20) (#4)⁠

May 19

fix: default enable_thinking to ON by adding else branch for <think> injection

fix: update generation_config.json to match project spec

fix: set pad_token_id to 151643 (endoftext)

Fix: restore Qwen3 chat template

Fix: restore Qwen3 chat template

Automated MNLP evaluation report (2026-05-19) (#3)⁠

May 18

Fix generation_config: correct pad_token_id, add bos_token_id, remove training-time overrides

Automated MNLP evaluation report (2026-05-18) (#2)⁠

May 17

Remove training_args.bin (training artifact, not needed for inference)

Remove trainer_state.json (training artifact, not needed for inference)

Remove scheduler.pt (training artifact, not needed for inference)

Remove rng_state.pth (training artifact, not needed for inference)

Remove optimizer.pt (training artifact, not needed for inference)

Distilled checkpoint (Qwen3-8B teacher, MATH train, 2 epochs)

May 13

Automated MNLP evaluation report (2026-05-13) (#1)⁠

May 12

Add thinking_token_budget=1500 to generation config

Fix system prompt with correct boxed format

Add default system prompt for boxed format compliance

Remove training_args.bin artifact

Restore max_new_tokens in generation_config

Remove training artifacts and checkpoint folders

Upload folder using huggingface_hub

Upload tokenizer

Upload Qwen3ForCausalLM

initial commit