Datasets:
The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
gatesv: struct<prompt: string, raw_prompt: string, samples: list<item: struct<sample_id: int64, raw_response (... 78 chars omitted)
child 0, prompt: string
child 1, raw_prompt: string
child 2, samples: list<item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, function (... 25 chars omitted)
child 0, item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, functional: bool, e (... 13 chars omitted)
child 0, sample_id: int64
child 1, raw_response: string
child 2, completion: string
child 3, syntax: bool
child 4, functional: bool
child 5, error: string
rotate100: struct<prompt: string, raw_prompt: string, samples: list<item: struct<sample_id: int64, raw_response (... 76 chars omitted)
child 0, prompt: string
child 1, raw_prompt: string
child 2, samples: list<item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, function (... 23 chars omitted)
child 0, item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, functional: bool, e (... 11 chars omitted)
child 0, sample_id: int64
child 1, raw_response: string
child 2, completion: string
child 3, syntax: bool
child 4, functional: bool
child 5, error: null
review2015_fsmonehot: struct<prompt: string, raw_prompt: string, samples: list<item: struct<sample_id: int64, raw_r
...
d: int64, raw_response (... 78 chars omitted)
child 0, prompt: string
child 1, raw_prompt: string
child 2, samples: list<item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, function (... 25 chars omitted)
child 0, item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, functional: bool, e (... 13 chars omitted)
child 0, sample_id: int64
child 1, raw_response: string
child 2, completion: string
child 3, syntax: bool
child 4, functional: bool
child 5, error: string
aggregate: struct<pass@1: double, pass@5: double, pass@10: double, syntax: double>
child 0, pass@1: double
child 1, pass@5: double
child 2, pass@10: double
child 3, syntax: double
timestamp: string
top_p: double
max_tokens: int64
temperature: double
engine: string
n_samples: int64
model: string
per_problem: list<item: struct<task_id: string, n: int64, syntax_passes: int64, func_passes: int64, syntax_rate: (... 57 chars omitted)
child 0, item: struct<task_id: string, n: int64, syntax_passes: int64, func_passes: int64, syntax_rate: double, pas (... 45 chars omitted)
child 0, task_id: string
child 1, n: int64
child 2, syntax_passes: int64
child 3, func_passes: int64
child 4, syntax_rate: double
child 5, pass@1: double
child 6, pass@5: double
child 7, pass@10: double
benchmark: string
benchmark_version: string
model_tag: string
to
{'model': Value('string'), 'model_tag': Value('string'), 'engine': Value('string'), 'benchmark': Value('string'), 'benchmark_version': Value('string'), 'n_samples': Value('int64'), 'temperature': Value('float64'), 'top_p': Value('float64'), 'max_tokens': Value('int64'), 'timestamp': Value('string'), 'aggregate': {'pass@1': Value('float64'), 'pass@5': Value('float64'), 'pass@10': Value('float64'), 'syntax': Value('float64')}, 'per_problem': List({'task_id': Value('string'), 'n': Value('int64'), 'syntax_passes': Value('int64'), 'func_passes': Value('int64'), 'syntax_rate': Value('float64'), 'pass@1': Value('float64'), 'pass@5': Value('float64'), 'pass@10': Value('float64')})}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
return get_rows(
^^^^^^^^^
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/src/services/worker/src/worker/utils.py", line 77, in get_rows
rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2690, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2227, in __iter__
for key, pa_table in self._iter_arrow():
^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2251, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 494, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 384, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 295, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 128, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2281, in table_cast
return cast_table_to_schema(table, schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2227, in cast_table_to_schema
raise CastError(
datasets.table.CastError: Couldn't cast
gatesv: struct<prompt: string, raw_prompt: string, samples: list<item: struct<sample_id: int64, raw_response (... 78 chars omitted)
child 0, prompt: string
child 1, raw_prompt: string
child 2, samples: list<item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, function (... 25 chars omitted)
child 0, item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, functional: bool, e (... 13 chars omitted)
child 0, sample_id: int64
child 1, raw_response: string
child 2, completion: string
child 3, syntax: bool
child 4, functional: bool
child 5, error: string
rotate100: struct<prompt: string, raw_prompt: string, samples: list<item: struct<sample_id: int64, raw_response (... 76 chars omitted)
child 0, prompt: string
child 1, raw_prompt: string
child 2, samples: list<item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, function (... 23 chars omitted)
child 0, item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, functional: bool, e (... 11 chars omitted)
child 0, sample_id: int64
child 1, raw_response: string
child 2, completion: string
child 3, syntax: bool
child 4, functional: bool
child 5, error: null
review2015_fsmonehot: struct<prompt: string, raw_prompt: string, samples: list<item: struct<sample_id: int64, raw_r
...
d: int64, raw_response (... 78 chars omitted)
child 0, prompt: string
child 1, raw_prompt: string
child 2, samples: list<item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, function (... 25 chars omitted)
child 0, item: struct<sample_id: int64, raw_response: string, completion: string, syntax: bool, functional: bool, e (... 13 chars omitted)
child 0, sample_id: int64
child 1, raw_response: string
child 2, completion: string
child 3, syntax: bool
child 4, functional: bool
child 5, error: string
aggregate: struct<pass@1: double, pass@5: double, pass@10: double, syntax: double>
child 0, pass@1: double
child 1, pass@5: double
child 2, pass@10: double
child 3, syntax: double
timestamp: string
top_p: double
max_tokens: int64
temperature: double
engine: string
n_samples: int64
model: string
per_problem: list<item: struct<task_id: string, n: int64, syntax_passes: int64, func_passes: int64, syntax_rate: (... 57 chars omitted)
child 0, item: struct<task_id: string, n: int64, syntax_passes: int64, func_passes: int64, syntax_rate: double, pas (... 45 chars omitted)
child 0, task_id: string
child 1, n: int64
child 2, syntax_passes: int64
child 3, func_passes: int64
child 4, syntax_rate: double
child 5, pass@1: double
child 6, pass@5: double
child 7, pass@10: double
benchmark: string
benchmark_version: string
model_tag: string
to
{'model': Value('string'), 'model_tag': Value('string'), 'engine': Value('string'), 'benchmark': Value('string'), 'benchmark_version': Value('string'), 'n_samples': Value('int64'), 'temperature': Value('float64'), 'top_p': Value('float64'), 'max_tokens': Value('int64'), 'timestamp': Value('string'), 'aggregate': {'pass@1': Value('float64'), 'pass@5': Value('float64'), 'pass@10': Value('float64'), 'syntax': Value('float64')}, 'per_problem': List({'task_id': Value('string'), 'n': Value('int64'), 'syntax_passes': Value('int64'), 'func_passes': Value('int64'), 'syntax_rate': Value('float64'), 'pass@1': Value('float64'), 'pass@5': Value('float64'), 'pass@10': Value('float64')})}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
VerilogEval v1 Results — yang-z/CodeV-QW-7B (Official CodeV Paper Fine-tune)
This dataset contains full benchmark results (aggregate scores + per-sample prompts and model responses) for evaluating the official CodeV paper fine-tune yang-z/CodeV-QW-7B on the VerilogEval v1 benchmark.
This is an independent reproduction of Table III of the CodeV paper (arXiv:2407.10424) using a fresh evaluation pipeline. All 6 metrics reproduce within ~1 point of the paper; 5/6 match or exceed paper values.
Companion datasets from the same pipeline:
muratkarahan/verilogeval-v1-qwen-base-results— Qwen2.5-Coder-7B basemuratkarahan/verilogeval-v1-codev-qwen-v2-results— my own CodeV-style fine-tune on Qwen2.5-Coder-7B base
Run Configuration
| Setting | Value |
|---|---|
| Model | yang-z/CodeV-QW-7B (Qwen2-7B base, CodeV Magicoder-style fine-tune) |
| Engine | vLLM 0.19.1, tensor_parallel_size=4, dtype=bfloat16, max_model_len=4096 |
| Hardware | 4× NVIDIA H100 80GB HBM3 (RunPod) |
| Benchmarks | VerilogEval-v1-Machine (143 problems) + VerilogEval-v1-Human (156 problems) |
| Temperatures | 0.2, 0.5, 0.8 |
| n_samples | 20 per problem |
| top_p | 0.95 |
| max_tokens | 2048 |
| Stop tokens | ["endmodule", "\n\include", "\n```"]` |
| Prompt format | Magicoder instruct wrapper (see below) |
| Simulator | Icarus Verilog 11.0 (iverilog -g2012 -s tb, then vvp -n) |
| Functional metric | Mismatches: 0 in N samples in vvp stdout |
Critical: Magicoder Prompt Wrapper
CodeV models are trained using the Magicoder training recipe, NOT raw code completion. Using the raw completion format that works for base Qwen or for CodeV-style raw-completion fine-tunes results in syntax_rate ~2% on this model, because the model re-emits the module top_module(...); header instead of only the body — causing a duplicate module definition at compile time.
The correct template (taken from the official CodeV inference script):
MAGICODER_PROMPT = """You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.
@@ Instruction
{instruction}
@@ Response
{response}"""
# instruction = detail_description + "\n" + module_header
# response = "" at inference time
The completion post-processing strips any regenerated module top_module (...); header from the output before splicing it with the testbench.
Aggregate Results
All values are mean across problems (in %). Best temperature per metric is bold.
VerilogEval-v1-Machine (143 problems)
| Temp | pass@1 | pass@5 | pass@10 | syntax |
|---|---|---|---|---|
| 0.2 | 79.2 | 83.8 | 85.1 | 97.8 |
| 0.5 | 79.4 | 87.5 | 89.7 | 97.4 |
| 0.8 | 76.7 | 88.2 | 91.1 | 96.5 |
VerilogEval-v1-Human (156 problems)
| Temp | pass@1 | pass@5 | pass@10 | syntax |
|---|---|---|---|---|
| 0.2 | 53.6 | 61.7 | 64.2 | 94.4 |
| 0.5 | 51.8 | 65.6 | 69.7 | 93.4 |
| 0.8 | 50.1 | 65.7 | 69.5 | 92.0 |
Per-metric best vs Paper (CodeV-QC, Table III)
| Machine p@1 | Machine p@5 | Machine p@10 | Human p@1 | Human p@5 | Human p@10 | |
|---|---|---|---|---|---|---|
| This run | 79.4 | 88.2 | 91.1 | 53.6 | 65.7 | 69.7 |
| Paper (CodeV-QC) | 80.1 | 87.9 | 90.5 | 53.2 | 65.1 | 68.5 |
| Δ | −0.7 | +0.3 | +0.6 | +0.4 | +0.6 | +1.2 |
5/6 metrics match or exceed the paper; all 6 within ~1 point. The CodeV paper Table III results are faithfully reproduced on a completely independent pipeline.
File Layout
Each of the 6 runs produces two files:
codev-qw-7b_{machine,human}_t{0.2,0.5,0.8}_<timestamp>.json <- aggregate scores + per-problem metrics
codev-qw-7b_{machine,human}_t{0.2,0.5,0.8}_<timestamp>_responses.json <- prompts + all raw completions
Aggregate file structure (*_<timestamp>.json)
{
"model": "yang-z/CodeV-QW-7B",
"model_tag": "codev-qw-7b",
"engine": "vLLM",
"benchmark": "VerilogEval-v1-machine",
"benchmark_version": "v1.0.0-fixed",
"n_samples": 20,
"temperature": 0.5,
"top_p": 0.95,
"max_tokens": 2048,
"timestamp": "20260420_112526",
"aggregate": {
"pass@1": 79.4, "pass@5": 87.5, "pass@10": 89.7, "syntax": 97.4
},
"per_problem": [ ... ]
}
Responses file structure (*_responses.json)
Keyed by task_id. Each entry contains both the full Magicoder-wrapped prompt sent to the model (prompt) and the raw description+header form used for testbench splicing (raw_prompt), plus all 20 completions with per-sample evaluation:
{
"mux2to1v": {
"prompt": "You are an exceptionally intelligent coding assistant...\n\n@@ Instruction\nThis module is a simple multiplexer...\nmodule top_module (\n\tinput [99:0] a,\n\t...\n);\n\n@@ Response\n",
"raw_prompt": "// This module is a simple multiplexer...\n\nmodule top_module (\n\tinput [99:0] a,\n\t...\n);",
"samples": [
{
"sample_id": 0,
"raw_response": "module top_module (\n\tinput [99:0] a,\n\t...\n);\n\tassign out = sel ? b : a;\n\nendmodule",
"completion": "\n\tassign out = sel ? b : a;\n\nendmodule",
"syntax": true,
"functional": true,
"error": null
},
...
]
},
...
}
raw_response shows what the model actually emitted (often with a regenerated header); completion shows what was spliced into the testbench after post-processing.
Reproducing
Source code: DerinVural/DSAI_BENCHMARK (see run_benchmark_v1_vllm_fixed.py and run_full_runpod_codev_qw.sh).
python3 run_benchmark_v1_vllm_fixed.py \
--model yang-z/CodeV-QW-7B --model_tag codev-qw-7b \
--benchmark machine --temperature 0.2 --n_samples 20 \
--tensor_parallel_size 4 --dtype bfloat16 \
--prompt_style magicoder \
--output_dir results_v1_fixed
Citation
@article{zhao2024codev,
title={CodeV: Empowering LLMs with HDL Generation through Multi-Level Summarization},
author={Zhao, Yang and others},
journal={arXiv preprint arXiv:2407.10424},
year={2024}
}
@inproceedings{liu2023verilogeval,
title={VerilogEval: Evaluating Large Language Models for Verilog Code Generation},
author={Liu, Mingjie and others},
booktitle={ICCAD},
year={2023}
}
- Downloads last month
- 69