noeme/Lura-1.2-500m

Lura 1.2 (500M)

Lura 1.2 is an aligned 500M parameter model fine-tuned using Online Reinforcement Learning from Human Feedback (RLHF) on top of SEN-AGI/Lura-1.1-500m.

Lura 1.2 transitions from unstructured single-line completions to clean, step-by-step chain-of-thought deductions, delivering measurable performance gains across math word problems and strict instruction following.

---

Benchmark Evaluation (100 Samples Each)

All evaluations were conducted using deterministic greedy decoding (do_sample=False, temperature=0.0) across 100 real samples per benchmark suite.

Summary Results Table

Benchmark Suite	Samples Tested	Lura 1.1 (Base)	Lura 1.2 (Main Model)	Net Gain
SVAMP (Elementary Math)	100	50.0%	55.0%	+5.0%
GSM8k (Multi-Step Math)	100	24.0%	30.0%	+6.0%
Instruction Following	103	36.9%	45.6%	+8.7%
Internal Mixed Benchmark	100	42.0%	54.0%	+12.0%

---

10 Showcase Reasoning Examples

1. Multi-Step Multiplication

Prompt: James decides to run 3 sprints 3 times a week. He runs 60 meters each sprint. How many total meters does he run a week?
Lura 1.2 Output:

James runs 3 sprints 3 times a week, so he runs 3 × 3 = 9 sprints in a week.
He runs 60 meters each sprint. Therefore, the total meters run in a week is:
9 × 60 = 540 meters

540

---

2. Multi-Tier Relational Reasoning

Prompt: Toulouse has twice as many sheep as Charleston. Charleston has 4 times as many sheep as Seattle. How many sheep do Toulouse, Charleston, and Seattle have together if Seattle has 20 sheep?
Lura 1.2 Output:

Seattle has 20 sheep.
Charleston has 4 times as many sheep as Seattle: 4 × 20 = 80 sheep.
Toulouse has twice as many sheep as Charleston: 2 × 80 = 160 sheep.
Total sheep: 20 + 80 + 160 = 260 sheep.

260

---

3. Proportional Age Derivation

Prompt: Brandon's iPhone is four times as old as Ben's iPhone. Ben's iPhone is two times older than Suzy's iPhone. If Suzy’s iPhone is 1 year old, how old is Brandon’s iPhone?
Lura 1.2 Output:

Suzy's iPhone is 1 year old.
Ben's iPhone is two times Suzy's iPhone: 2 × 1 = 2 years old.
Brandon's iPhone is four times Ben's iPhone: 4 × 2 = 8 years old.

8

---

4. Partitioning & Subtraction

Prompt: Grandma Jones baked 5 apple pies for the fireman's luncheon. She cut each pie into 8 pieces and set them out. At the end of the evening, there were 14 pieces remaining. How many pieces were taken by the guests?
Lura 1.2 Output:

Grandma Jones baked 5 pies, cut each into 8 pieces, so the total number of pieces is:
5 × 8 = 40 pieces
At the end of the evening, there were 14 pieces remaining. So the number of pieces taken by the guests is:
40 - 14 = 26 pieces

26

---

5. Multi-Item Cluster Summation

Prompt: A raspberry bush has 6 clusters of 20 fruit each and 67 individual fruit scattered across the bush. How many raspberries are there total?
Lura 1.2 Output:

The raspberry bush has 6 clusters of 20 fruit each:
Total fruit = 6 × 20 = 120
67 individual fruit are scattered across the bush.
Total raspberries = 120 + 67 = 187

187

---

6. Intermediate Step Deduction

Prompt: Marco and his dad went strawberry picking. Together they collected strawberries that weighed 22 pounds. On the way back Marco's dad found 30 more pounds. Marco's strawberries now weighed 36 pounds. How much did his dad's strawberries weigh now?
Lura 1.2 Output:

Marco and his dad together collected 22 pounds.
On the way back, he found 30 more pounds. So his dad's strawberries now weigh:
22 + 30 = 52 pounds
Marco's strawberries now weigh 36 pounds. The difference is the amount his dad found:
52 - 36 = 16 pounds
His dad's strawberries weigh 16 pounds.

---

7. Multi-Group Addition & Net Difference

Prompt: In a school there are 362 boys and 257 girls. 403 more girls joined the school. How many more girls than boys does the school have?
Lura 1.2 Output:

You start with 362 boys and 257 girls.
403 more girls join the school: 257 + 403 = 660
To find how many more girls than boys:
660 - 362 = 298

298

---

8. Budget & Multi-Step Purchases

Prompt: Emma had $100. She bought 3 books for $12 each and a backpack for $30. How much money does she have left?
Lura 1.2 Output:

Emma starts with $100.

• She buys 3 books at $12 each: 3 × 12 = $36
• She buys a backpack for $30
  Total money spent = 36 + 30 = $66
  Money left = 100 - 66 = $34

34

---

9. Rate & Distance Consumption

Prompt: A car tank holds 50 liters of fuel. The car uses 8 liters per 100 km. How many liters remain after driving 300 km?
Lura 1.2 Output:

The car uses 8 liters per 100 km.
For 300 km, the fuel used is:
(300 ÷ 100) × 8 = 3 × 8 = 24 liters
The tank initially holds 50 liters.
Remaining fuel after 300 km:
50 - 24 = 26 liters

26

---

10. Discount & Price Calculation

Prompt: Each pack of DVDs costs $76. If there is a discount of $25 on each pack, how much do you have to pay to buy each pack?
Lura 1.2 Output:

Each pack costs $76.
You get a 25-dollar discount on each pack:
76 - 25 = 51 dollars

51

---

Quickstart & Usage

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "noeme/Lura-1.2-500m"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
    device_map="auto"
)

prompt = "A baker has 48 muffins. He sells 18 in the morning and 12 in the afternoon. How many muffins are left?"
messages = [{"role": "user", "content": prompt}]
formatted = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

inputs = tokenizer(formatted, return_tensors="pt").to(model.device)
with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=256,
        temperature=0.3,
        repetition_penalty=1.15,
        do_sample=True
    )

print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))