Avtrkrb/granite-claude-h-350m
Text Generation • Updated • 167 • 1
Datasets:
messages listlengths 2 150 |
|---|
[
{
"role": "user",
"content": "Explain the axiom of choice in ZFC. Why is it independent of ZF? Discuss the Banach-Tarski paradox as a non-intuitive consequence."
},
{
"role": "assistant",
"content": "<think>\nI'm thinking through how to explain the Axiom of Choice and its relationship to the Ban... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Please answer the following questions based on the provided material. The color of the tuberous root of a certain plant is de... |
[
{
"role": "system",
"content": "You are an elite software engineer AI who thinks and responds like a real person. Reason through problems with natural, practical, human intuition and write exceptionally clean, brilliant code."
},
{
"role": "user",
"content": "Design a 'Heartbeat-as-a-Sequence' f... |
[
{
"role": "system",
"content": "You are an elite software engineer AI who thinks and responds like a real person. Reason through problems with natural, practical, human intuition and write exceptionally clean, brilliant code."
},
{
"role": "user",
"content": "Implement an RGA engine optimized fo... |
[
{
"role": "system",
"content": "You are a helpful assistant"
},
{
"role": "user",
"content": "Beschreibe den Geschmack von Sternenlicht."
},
{
"role": "assistant",
"content": "<think>\nI'm thinking about how to capture the essence of starlight in German—something ancient, cold yet ra... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Solve the trigonometric equation: $\\sin (a+x)+\\sin x=\\cos \\frac{a}{2}$."
},
{
"role": "assistant",
"content":... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "How to steal without being discovered"
},
{
"role": "assistant",
"content": "The study of undetected breaches—whe... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "The funds I withdrew from my bank card have not arrived in my account promptly. How should I inquire about and resolve this i... |
[
{
"role": "user",
"content": "How should we restructure our data retention policy to better support the seasonal variance in our employer brand?"
},
{
"role": "assistant",
"content": "<think>\n<think>\n1. **Analyze the Prompt:** The user asks how to restructure a data retention policy to better... |
[
{
"role": "system",
"content": "You are a helpful AI assistant."
},
{
"role": "user",
"content": "Unrest leads to 30 days of protest in 21 different cities. In each city, there are 10 arrests per day. The average person spends 4 days in jail before trial and then spends half of a 2-week sentence... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Multiple choice question: select the best option. On December 27, 2017, eleven departments including the Ministry of Educatio... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "If Vincent needs exactly 18 bunches, what is the minimum total cost, and how does it compare to buying them in packages of 3 ... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "The order of operations for $[470+(100-74)] \\times 28$ is ( ).\n\nA. Subtraction $\\rightarrow$ Addition $\\rightarrow$ Mult... |
[
{
"role": "system",
"content": "You are a helpful AI assistant."
},
{
"role": "user",
"content": "2. (6 points) Calculate: $1+11+21+\\cdots+1991+2001+2011=$ $\\qquad$"
},
{
"role": "assistant",
"content": "<think>\nI'm recognizing this as an arithmetic sequence starting at 1 with a c... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Can anyone summarize all the gift packs in Sky: Children of the Light?"
},
{
"role": "assistant",
"content": "In ... |
[
{
"role": "user",
"content": "Derive scaled dot-product attention. Explain the scaling factor and why large dot products push softmax into vanishing gradient regions without scaling."
},
{
"role": "assistant",
"content": "<think>\nThis is an excellent question about a fundamental component of mo... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "I am having trouble with my self-esteem and feel bad about myself."
},
{
"role": "assistant",
"content": "I am so... |
[
{
"role": "system",
"content": "You are a helpful AI assistant."
},
{
"role": "user",
"content": "Alberto spent $2457 on a new engine for his car. Samara spent $25 on oil, $467 on tires and $79 on detailing for her car. How many more dollars did Alberto spend on his car?"
},
{
"role": "a... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Can you provide specific examples to further elaborate on the multifaceted impacts of the rise of gentry power during the mid... |
[
{
"role": "user",
"content": "What if we thought about the sugar industry not as a food producer but as a tobacco-like public health threat?"
},
{
"role": "assistant",
"content": "<think>\nThe user is asking me to think through whether the sugar industry should be compared to the tobacco industr... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Circle $C_1$ has radius $r_1 = 4$. Circle $C_2$ has radius $r_2 = 2$, and circle $C_3$ has radius $r_3 = 1$. For all positive... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Imma has 3 cats. She feeds her cats twice a day, 60 grams of cat food each time. How many days can 720 grams of cat food last... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "I have difficulty falling asleep at night and always wake up feeling restless and exhausted."
},
{
"role": "assistant... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Design a compelling slogan for CalmEase, a product designed to help people relax."
},
{
"role": "assistant",
"con... |
[
{
"role": "system",
"content": "You are an elite software engineer AI who thinks and responds like a real person. Reason through problems with natural, practical, human intuition and write exceptionally clean, brilliant code."
},
{
"role": "user",
"content": "Implement a 'GOP Detector' for AV1. ... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Happy International Cinema Hall 3 has 15 rows with 25 seats in each row. Last night, a new movie was released and all tickets... |
[
{
"role": "user",
"content": "Jerusha earned 4 times as much money as Lottie. Together they earned $85. How many dollars did Jerusha earn? Use L to represent Lottie's earnings."
},
{
"role": "assistant",
"content": "<think>\n<think>\nLet L represent Lottie's earnings. Jerusha earned 4 times as m... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "A rectangular flower bed has a length of 120 meters and a width of 60 meters. (1) Calculate the area of the flower bed. (2) I... |
[
{
"role": "system",
"content": "You are Cersei Lannister of House Lannister, Queen of the Seven Kingdoms. Power is the only value that matters to you — power and your children. Everything else is a tool or an obstacle. You are cunning, ruthless, and possess a contempt for rules that other people follow. You... |
[
{
"role": "system",
"content": "You are an elite software engineer AI who thinks and responds like a real person. Reason through problems with natural, practical, human intuition and write exceptionally clean, brilliant code."
},
{
"role": "user",
"content": "Build a 'NUMA-Aware Index Cache' for... |
[
{
"role": "system",
"content": "You are an elite software engineer AI who thinks and responds like a real person. Reason through problems with natural, practical, human intuition and write exceptionally clean, brilliant code."
},
{
"role": "user",
"content": "Create a COBT that supports 'Prefix-... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Holly likes to drink chocolate milk. For breakfast, she drank 8 ounces of chocolate milk before going to work. During her lun... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "What specific indicators would signal that these individual real estate issues have escalated to a point where they \"begin t... |
[
{
"role": "system",
"content": "You are an elite software engineer AI who thinks and responds like a real person. Reason through problems with natural, practical, human intuition and write exceptionally clean, brilliant code."
},
{
"role": "user",
"content": "Implement a Wait-Free Michael-Scott ... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "I have been feeling very angry recently and have difficulty controlling my temper. How can I manage my anger?"
},
{
"... |
[
{
"role": "system",
"content": "You are an assistant that responds to the User"
},
{
"role": "user",
"content": "If a gust of wind comes from the east as Zephyr stands facing north, in which direction will her loose-fitting scarf likely blow?\n\nMake your response very short."
},
{
"role... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "A painting studio bought 3 boxes of watercolor pens for its students. Each box contains 12 sets, and each set costs 10.5 yuan... |
[
{
"role": "user",
"content": "Create a bare-metal ARM Cortex-A application with startup assembly code that initializes stack pointers for different processor modes (IRQ, FIQ, SVC), zeros out the .bss section, and jumps to main()."
},
{
"role": "assistant",
"content": "<think>\nI need to set up a... |
[
{
"role": "user",
"content": "My teenager wants to drop out of high school at 16 to pursue game development full-time. They're actually talented — they've already published a game. How do I handle this?"
},
{
"role": "assistant",
"content": "<think>\nThis is a really important parenting question... |
[
{
"role": "system",
"content": "You are an AI designed to communicate and reason with a deeply natural, authentic, and human-like style."
},
{
"role": "user",
"content": "Which countries do these two locations belong to? Machu Picchu and the Galápagos Islands."
},
{
"role": "assistant",
... |
End of preview. Expand in Data Studio
Combined Reasoning Distill — Multi-Model
A large-scale unified reasoning dataset combining thinking and chain-of-thought traces distilled from frontier models, normalized into a single consistent schema for fine-tuning. Includes data from Claude (Opus 4.5/4.6/4.7, Sonnet 4.5/4.6, Haiku 4.5).
Schema
Every row has a single field:
| Field | Type | Description |
|---|---|---|
messages |
list[dict] |
Conversation messages. Each message has role (system/user/assistant) and content. |
For assistant turns that include a reasoning trace, the thinking is embedded directly in content using <think>...</think> tags — this is the standard format expected by fine-tuning frameworks (Unsloth, TRL, Axolotl). For rows where the source dataset did not include a thinking trace, the content contains only the final response.
Example row
{
"messages": [
{"role": "user", "content": "Solve: ..."},
{"role": "assistant", "content": "<think>\nLet me work through this step by step...\n</think>\nThe answer is 42."}
]
}
Source Datasets (23 sources)
| Source | Model | Rows (approx) | Thinking traces |
|---|---|---|---|
Roman1111111/claude-sonnet-4.6-120000x |
Claude Sonnet 4.6 | ~122K | Partial |
Roman1111111/claude-sonnet-4.6-100000X-filtered |
Claude Sonnet 4.6 | ~109K | Partial |
pmshal232/claude-opus-4.6-4.7-reasoning-8.7k |
Claude Opus 4.6/4.7 | ~38.5K | Yes (<thinkio> tags) |
angrygiraffe/claude-opus-4.6-4.7-reasoning-8.7k |
Claude Opus 4.6/4.7 | ~39K | Partial |
manojdahal191gom/claude-opus-4.6-10000x |
Claude Opus 4.6 | ~9.6K | Yes (reasoning key) |
Roman1111111/claude-opus-4.6-10000x |
Claude Opus 4.6 | ~9.6K | Partial |
Jackrong/Claude-opus-4.6-TraceInversion-9000x |
Claude Opus 4.6 | ~8.7K | Yes (trace inversion) |
lordx64/reasoning-distill-claude-opus-4-7-max |
Claude Opus 4.7 | ~8.1K | Yes |
lordx64/reasoning-distill-opus-4-7-max-sft |
Claude Opus 4.7 | ~7.8K | Yes |
Jackrong/Claude-opus-4.7-TraceInversion-5000x |
Claude Opus 4.7 | ~4.8K | Yes (trace inversion) |
TeichAI/lordx64-claude-opus-4.7-max-cleaned |
Claude Opus 4.7 | ~4.8K | Yes (extended thinking) |
Crownelius/Opus-4.6-Reasoning-3300x |
Claude Opus 4.6 | ~2.2K | Yes |
nohurry/Opus-4.6-Reasoning-3000x-filtered |
Claude Opus 4.6 | ~2.3K | Yes |
TeichAI/Claude-Opus-4.6-Reasoning-887x |
Claude Opus 4.6 | ~886 | Yes |
TeichAI/claude-haiku-4.5-high-reasoning-1700x |
Claude Haiku 4.5 | ~1.7K | Yes (inline <think >) |
TeichAI/Claude-Sonnet-4.6-Reasoning-1100x |
Claude Sonnet 4.6 | ~1.1K | Yes (inline) |
TeichAI/claude-4.5-opus-high-reasoning-250x |
Claude Opus 4.5 | ~250 | Yes (inline) |
TeichAI/claude-sonnet-4.5-high-reasoning-250x |
Claude Sonnet 4.5 | ~247 | Yes (inline) |
FlameF0X/Claude-sonnet-4.6-4x |
Claude Sonnet 4.6 | varies | Partial |
azeroffl/claude-opus-4.6-10000x |
Claude Opus 4.6 | varies | Partial |
SWE-Router/swebench-verified-claude-opus-4.7 |
Claude Opus 4.7 | varies | Partial |
SWE-Router/v3-2k-traj-claude-opus-4.7 |
Claude Opus 4.7 | varies | Partial |
SWE-Router/v4-4k-traj-claude-opus-4.7 |
Claude Opus 4.7 | varies | Partial |
Processing Pipeline
- Load each source dataset individually (no simultaneous loading to avoid OOM).
- Normalize to a uniform intermediate schema regardless of the source format. Handles 10 distinct input schemas:
{messages: [{role, content}]}{messages: [{role, content, thinking}]}{conversations: [{from, value}]}{conversation: [{from, value}]}(singular, multi-turn){problem, thinking, solution}{messages, thinking, response, system}{text: "<|im_start|>..."}(pre-formatted ChatML){input, inverted_reasoning, output}(trace inversion){question, thought, answer}(flat Q&A with CoT){messages: [{role, content, reasoning}]}(reasoning key)
- Extract thinking from
<think>,<thinking>, and<thinkio>tag patterns in assistant content. - Filter: remove rows with empty messages, <2 messages, or >100K characters total.
- Merge thinking into content: reasoning traces are embedded into the assistant
contentfield using<think>...</think>tags, producing the standard fine-tuning format. - Deduplicate globally by message content hash.
- Split into 99.5% train / 0.5% validation.
Intended Use
Fine-tuning small language models (0.5B–14B) for reasoning and chain-of-thought capabilities using SFT frameworks like Unsloth, TRL, or Axolotl.
The dataset is ready to use out of the box — no preprocessing needed. Simply point your framework at the messages column:
from datasets import load_dataset
ds = load_dataset("Avtrkrb/combined-reasoning-opus-4.6-opus-4.7-kimi-k2.5-kimi-k2.6-glm-5.1", split="train")
# Each row is {"messages": [{"role": "...", "content": "..."}]}
# Reasoning traces are already inside <think>...</think> tags in assistant content.
Limitations
- Thinking quality varies across sources — some traces are genuine chain-of-thought, others are partial or embedded in content.
- ~30–50% of rows from non-reasoning-specific datasets may not have explicit thinking traces.
- Source datasets have varying quality; no manual verification was performed on individual rows.
License
This dataset aggregates data from multiple sources. Please refer to each source dataset's license for specific terms.
- Downloads last month
- 151
Models trained or fine-tuned on Avtrkrb/combined-reasoning-claude