Datasets:
prompt_id large_string | prompt large_string | source large_string | input_category large_string | test_status large_string | primary_suite large_string | metadata list | test_suites list | sample_id string | generation_index int64 | candidate_code string | reasoning_content string | generation string | candidate_validation_status string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
algebra_discovery_00c1adbaf6d236d28c78c772 | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nFor each b(...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_00c1adbaf6d236d28c78c772:0 | 0 | "```python\nimport torch\nfrom torch.utils.cpp_extension import load_inline\n\n_CPP_SRC = r'''\n#inc(...TRUNCATED) | "We need answer user's request. Need implement CUDA kernel module. Need think carefully. Need produc(...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"792868acfda47322724b01bcfeaeadaa2(...TRUNCATED) | not_evaluated |
algebra_discovery_0187ef8576247c412d46964d | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nApply a To(...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_0187ef8576247c412d46964d:0 | 0 | "import torch\nfrom torch.utils.cpp_extension import load_inline\n\n_CPP_SRC = r'''\n#include <torch(...TRUNCATED) | "We need answer user's request. Need implement self-contained Python module exposing solve(c, r, X, (...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"f2cba9449c003d6ef4dc99726da27e0b0(...TRUNCATED) | not_evaluated |
algebra_discovery_01f80bdafca3f48274dfd502 | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nApply the (...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_01f80bdafca3f48274dfd502:0 | 0 | "import threading\nimport torch\nfrom torch.utils.cpp_extension import load_inline\n\n_CPP_SRC = r\"(...TRUNCATED) | "We need answer user's request. Need implement self-contained Python module exposing solve(A,B,X,*,d(...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"3569ef5262a6a68ec1247bda0ba35e5e8(...TRUNCATED) | not_evaluated |
algebra_discovery_058f1ba45fabab72656ccbc5 | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nApply the (...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_058f1ba45fabab72656ccbc5:0 | 0 | "```python\nimport torch\nfrom torch.utils.cpp_extension import load_inline\n\n_CPP_SRC = r'''\n#inc(...TRUNCATED) | "We need answer user's request. Need implement CUDA kernels via PyTorch bindings, self-contained Pyt(...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"03dfaeb52d282a89498a7dc61153307e8(...TRUNCATED) | not_evaluated |
algebra_discovery_06f7f8879f2fb1014c336887 | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nApply an N(...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_06f7f8879f2fb1014c336887:0 | 0 | "```python\nimport torch\nfrom torch.utils.cpp_extension import load_inline\n\n\n_CPP_SRC = r\"\"\"\(...TRUNCATED) | "We need answer user's request. Need implement CUDA kernel via PyTorch bindings, self-contained Pyth(...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"2593109c1e7f04d0520e47c7bd262113d(...TRUNCATED) | not_evaluated |
algebra_discovery_07915c9c1e120535f829d689 | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nEvaluate t(...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_07915c9c1e120535f829d689:0 | 0 | "```python\nimport threading\nimport torch\nfrom torch.utils.cpp_extension import load_inline\n\n__a(...TRUNCATED) | "We need answer user's request. Need implement self-contained Python module exposing solve(band, x, (...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"a1bbcf5059998121ef68885e1f76cbed7(...TRUNCATED) | not_evaluated |
algebra_discovery_07ce7d26d5e85d3723ce5f8f | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nApply the (...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_07ce7d26d5e85d3723ce5f8f:0 | 0 | "import torch\nfrom torch.utils.cpp_extension import load_inline\n\n_CPP = r'''\n#include <torch/ext(...TRUNCATED) | "We need answer user's request. Need implement CUDA kernel module. Need think carefully. User wants (...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"23cd9589af56e8cd1cdc0c2187f36d9a2(...TRUNCATED) | not_evaluated |
algebra_discovery_07da416519219b65817f5edc | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nMatrix-vec(...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_07da416519219b65817f5edc:0 | 0 | "```python\nimport torch\nfrom torch.utils.cpp_extension import load_inline\n\n_CPP = (\n \"torch(...TRUNCATED) | "We need answer user's request. Need implement CUDA kernel module. We need think carefully. User wan(...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"d59f596eb94ef7425974ffe16544eeeeb(...TRUNCATED) | not_evaluated |
algebra_discovery_08afa92ae78d37b11db1ba57 | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nCompute th(...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_08afa92ae78d37b11db1ba57:0 | 0 | "```python\nimport torch\nfrom torch.utils.cpp_extension import load_inline\n\n_CPP_SRC = r'''\n#inc(...TRUNCATED) | "We need answer user's request. Need implement self-contained Python module exposing solve(c, r, X, (...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"d514eee5a3ad0f5e385c1d8630edd2c18(...TRUNCATED) | not_evaluated |
algebra_discovery_0aa7109ada420037e4606cf0 | "Implement the following mathematical computation efficiently using custom CUDA kernels.\nCompute th(...TRUNCATED) | glm53_algebra_discovery | algebra_matrix | ready | functional | [{"filename":"original_prompt.json","content":"{\"algebra_relaxation\":{\"cosine_threshold\":0.97,\"(...TRUNCATED) | [{"suite_id":"functional","type":"functional","coverage_scope":"numerical_outputs_with_task_specific(...TRUNCATED) | algebra_discovery_0aa7109ada420037e4606cf0:0 | 0 | "import torch\nfrom torch.utils.cpp_extension import load_inline\n\n_CPP_SRC = r'''\n#include <torch(...TRUNCATED) | "We need answer user's request. Need implement self-contained Python module exposing solve(alpha, A,(...TRUNCATED) | "{\"finish_reason\": \"stop\", \"hashes\": {\"messages_sha256\": \"c45230fa0dcdf6a8604b69b8801161176(...TRUNCATED) | not_evaluated |
End of preview. Expand in Data Studio
v7 one-shot candidate subset
Complete successful saved responses joined to their original ready v7 rows.
Original v7 columns and embedded test assets are retained. Development path prefixes in metadata are normalized to artifact:// references; embedded content hashes are regenerated. Prompts, candidate answers and reasoning are unchanged.
test_status: ready describes the original test availability, NOT candidate correctness.
No generated code or test suite was executed. No new API calls were made.
Added columns: sample_id, generation_index, candidate_code (verbatim final
answer, potentially including Markdown), reasoning_content, generation (JSON
string with exact request, usage, hashes and raw-record byte offsets), and
candidate_validation_status (always not_evaluated). Code presence is not
individually certified; incomplete/failed responses are excluded. No partial code
is promoted to a complete candidate. Reasoning is separate from candidate text.
from datasets import load_dataset
ds = load_dataset('Miaow-Lab/v7-kernel-candidates', split='train')
print(ds[0]['candidate_code'])
Equivalent rows are in exports/prompts.jsonl.gz. Raw SSE is retained in the
original collection JSONL, referenced by SHA-256 and byte offset, rather than
duplicated here. Preserve that source for full auditing. This is an unvalidated research subset, not the full 17,958-task ready set. Source-specific licenses and attribution in inherited metadata apply; no uniform permissive license is asserted. artifact:// references identify provenance and are not runtime filesystem paths.
Collection
5,164 complete responses from Qwen3.8-Flash via Paracloud. Failed and incomplete responses are excluded. Most full-run requests used max_tokens=131072; earlier pilot samples used 32768. See each row for exact settings. The subset reflects interrupted collection, not random sampling. Generated answers can be incorrect, violate task requirements, contain Markdown or use PyTorch fallbacks. No correctness or performance claims are made. Raw transport records are retained locally and are not included in this public release.
- Downloads last month
- 53