Datasets:

Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
35
101
categories
listlengths
2
2
input
dict
output
dict
harness
dict
ttsky26b-5iri-tt-riscv-cpu-task_vga_sync_0001
[ "cid002", "medium" ]
{ "prompt": "You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. You are given the full source code of a Verilog project. One of the modules in the project is incomplete — its implementation is missing and marked with:\n\n// >>> Module Implementation Begin\n\n// <<< Module Implemen...
{ "response": "", "context": { "rtl/generated.v": " // 640x480 @ 60Hz, 25MHz pixel clock\n localparam H_ACTIVE = 640;\n localparam H_FP = 16;\n localparam H_SYNC = 96;\n localparam H_BP = 48;\n localparam H_TOTAL = H_ACTIVE + H_FP + H_SYNC + H_BP;\n\n localparam V_ACTIVE = 480;\n...
{ "files": { "Dockerfile": "FROM __OSS_SIM_IMAGE__", "src/golden_solution.v": "`line 1 \"tt_um_systolic_vga.v\" 1\n// Tiny Tapeout top: 8x8 ternary systolic-array demo rendered directly to VGA.\n//\n// The logical matrix is 8x8, but it is computed in two passes using a 4x8\n// systolic slice so the hardware o...
ttsky26b-5iri-tt-riscv-gpu-task_spi_slave_0002
[ "cid002", "medium" ]
{ "prompt": "You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. You are given the full source code of a Verilog project. One of the modules in the project is incomplete — its implementation is missing and marked with:\n\n// >>> Module Implementation Begin\n\n// <<< Module Implemen...
{ "response": "", "context": { "rtl/generated.v": "\n // --- Synchronizers (2-FF for metastability) ---\n reg [2:0] sclk_sync;\n reg [2:0] cs_n_sync;\n reg [1:0] mosi_sync;\n\n wire sclk_rise = (sclk_sync[2:1] == 2'b01);\n wire sclk_fall = (sclk_sync[2:1] == 2'b10);\n wire cs_active = ~cs_n...
{ "files": { "Dockerfile": "FROM __OSS_SIM_IMAGE__", "src/golden_solution.v": "`line 1 \"tt_um_riscv_gpu.v\" 1\n`timescale 1ns/1ps\n`default_nettype none\n\n// TinyTapeout top wrapper for BitNet b1.58-style 4x4 matrix multiply accelerator (W1.58A8).\n// Weights (B) are ternary {-1, 0, +1}; activations (A) are...
ttihp26a-ADDTDR-tt-pwm-controller-task_PWM_0003
[ "cid002", "medium" ]
{ "prompt": "You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. You are given the full source code of a Verilog project. One of the modules in the project is incomplete — its implementation is missing and marked with:\n\n// >>> Module Implementation Begin\n\n// <<< Module Implemen...
{ "response": "", "context": { "rtl/generated.v": "\n reg [8:0] PWM_accumulator;\n\n initial\n PWM_accumulator = 0;\n\n always @(posedge clk) begin\n PWM_accumulator <= PWM_accumulator[7:0] + PWM_in;\n end\n assign PWM_out = PWM_accumulator[8];\n" } }
{ "files": { "Dockerfile": "FROM __OSS_SIM_IMAGE__", "src/golden_solution.v": "`line 1 \"project.v\" 1\n/*\n * Copyright (c) 2024 Your Name\n * SPDX-License-Identifier: Apache-2.0\n */\n\n`default_nettype none\n\nmodule tt_um_pwm_controller_atudoroi (\n input wire [7:0] ui_in, // Dedicated inputs\n ...
ttihp26a-ADDTDR-tt-pwm-controller-task_main_0004
[ "cid002", "medium" ]
{"prompt":"You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. Y(...TRUNCATED)
{"response":"","context":{"rtl/generated.v":"\n\n\twire tx_busy;\n\twire RxD_data_ready;\n\twire [7:(...TRUNCATED)
{"files":{"Dockerfile":"FROM __OSS_SIM_IMAGE__","src/golden_solution.v":"`line 1 \"project.v\" 1\n/*(...TRUNCATED)
ttihp26a-ADDTDR-tt-pwm-controller-task_uart_transmitter_0005
[ "cid002", "medium" ]
{"prompt":"You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. Y(...TRUNCATED)
{"response":"","context":{"rtl/generated.v":"\n\tparameter ClkFrequency = 12000000;\t// 12MHz\n\tpar(...TRUNCATED)
{"files":{"Dockerfile":"FROM __OSS_SIM_IMAGE__","src/golden_solution.v":"`line 1 \"project.v\" 1\n/*(...TRUNCATED)
ttsky25b-AG2048-tiny-calculator-asic-task_button_reader_0006
[ "cid002", "medium" ]
{"prompt":"You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. Y(...TRUNCATED)
{"response":"","context":{"rtl/generated.v":" /* \n Button reader module, scans a 4x4 button mat(...TRUNCATED)
{"files":{"Dockerfile":"FROM __OSS_SIM_IMAGE__","src/golden_solution.v":"`line 1 \"project.v\" 1\n/*(...TRUNCATED)
ttsky26a-Acrazt05-Titan-Processing-Unit-TPU-task_alu_0007
[ "cid002", "medium" ]
{"prompt":"You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. Y(...TRUNCATED)
{"response":"","context":{"rtl/generated.v":"\n wire [15:0] inter_x_1 = zx ? 16'd0 : x; \n wir(...TRUNCATED)
{"files":{"Dockerfile":"FROM __OSS_SIM_IMAGE__","src/golden_solution.v":"`line 1 \"project.v\" 1\n/*(...TRUNCATED)
ttsky26a-Acrazt05-Titan-Processing-Unit-TPU-task_computer_0008
[ "cid002", "medium" ]
{"prompt":"You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. Y(...TRUNCATED)
{"response":"","context":{"rtl/generated.v":"\n wire [7:0] usable_uio_oe;\n assign uio_oe = {u(...TRUNCATED)
{"files":{"Dockerfile":"FROM __OSS_SIM_IMAGE__","src/golden_solution.v":"`line 1 \"project.v\" 1\n/*(...TRUNCATED)
ttsky26a-Acrazt05-Titan-Processing-Unit-TPU-task_cpu_0009
[ "cid002", "medium" ]
{"prompt":"You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. Y(...TRUNCATED)
{"response":"","context":{"rtl/generated.v":"\n wire isCInstruction = instruction[15];\n\n wir(...TRUNCATED)
{"files":{"Dockerfile":"FROM __OSS_SIM_IMAGE__","src/golden_solution.v":"`line 1 \"project.v\" 1\n/*(...TRUNCATED)
ttsky26a-Acrazt05-Titan-Processing-Unit-TPU-task_gpio_0010
[ "cid002", "medium" ]
{"prompt":"You are an expert Verilog hardware designer working on a complete FPGA or ASIC project. Y(...TRUNCATED)
{"response":"","context":{"rtl/generated.v":"\n reg [7:0] outputs_reg;\n\n always @(posedge cl(...TRUNCATED)
{"files":{"Dockerfile":"FROM __OSS_SIM_IMAGE__","src/golden_solution.v":"`line 1 \"project.v\" 1\n/*(...TRUNCATED)
End of preview. Expand in Data Studio

This is a version of NotSoTiny-26-07 benchmark modified to work with Nvidia's CVDP framework

If you plan to run this benchmark via CVDP framework, proceed with this version. Otherwise refer to the main dataset for additional info: HPAI-BSC/NotSoTiny-26-07

Subsets and Shuttles

The default dataset contains all NotSoTiny-26-07 Tiny Tapeout shuttles combined onto a single dataset comprising 1128 total tasks. However, you can also access individual shuttles as subsets.

Subset # Tasks Launched date Tiny Tapeout source
default 1,128 - -
ttsky25b 140 2025-09-18 https://tinytapeout.com/chips/ttsky25b/
ttgf0p2 46 2025-11-06 https://tinytapeout.com/chips/ttgf0p2/
ttihp26a 191 2025-11-25 https://tinytapeout.com/chips/ttihp26a/
ttsky26a 233 2026-02-27 https://tinytapeout.com/chips/ttsky26a/
ttihp0p4 12 2026-03-27 https://tinytapeout.com/chips/ttihp0p4/
ttgf26a 84 2026-04-17 https://tinytapeout.com/chips/ttgf26a/
ttsky26b 328 2026-04-25 https://tinytapeout.com/chips/ttsky26b/
ttgf0p3 3 2026-06-01 https://tinytapeout.com/chips/ttgf0p3/
ttgf26b 91 2026-06-05 https://tinytapeout.com/chips/ttgf26b/

Run it with CVDP

1. Clone the CVDP framework:

$ git clone https://github.com/NVlabs/cvdp_benchmark.git
$ cd cvdp_benchmark/
$ uv init && uv add -r requirements.txt

2. Download the dataset shuttles:

$ git lfs install
$ git clone https://huggingface.co/datasets/HPAI-BSC/NotSoTiny-26-07-CVDP
$ ls NotSoTiny-26-07-CVDP/shuttles/
    ttsky25b.jsonl   ttgf0p2.jsonl   ttihp26a.jsonl   ttsky26a.jsonl   ttihp0p4.jsonl   ttgf26a.jsonl   ttsky26b.jsonl   ttgf0p3.jsonl   ttgf26b.jsonl
$ cp .env.example .env && echo "OSS_SIM_IMAGE=ggcr0/turtle-eval:2.3.4" >> .env

3. [Optional] Validate with golden solutions:

$ uv run run_benchmark.py -f NotSoTiny-26-07-CVDP/shuttles/ttsky25b.jsonl

4. Run the benchmark (inference + eval):

$ export OPENROUTER_API_KEY=sk-or-v1-... 
$ uv run run_samples.py \
       -f NotSoTiny-26-07-CVDP/shuttles/ttsky25b.jsonl \
       -l \
       -m mistralai/codestral-2508 \
       -c examples/openrouter_factory.py \
       -n 5 \
       -k 1

You can repeat this process for any other shuttle present in the shuttles/ dir.

We also recommend the usage of the -t <workers> flag to speed-up the process.

This CVDP version contains 1,128 of the 1,146 tasks in NotSoTiny-26-07. 18 tasks from 3 projects were removed because their golden solution does not pass the Icarus Verilog syntax check that the CVDP harness runs before equivalence checking, due to issues in the source design:

  • EzraWolf-TinyMOA-IHP26a (ttihp26a, 5 tasks): the aggregated design is missing the tinymoa_decoder module
  • FHW-Appel-TTgf26a-spiPWMio (ttgf26a, 8 tasks): the cocotb testbench helper cocotb_iverilog_dump was merged into the design
  • elvtide01-7SegmentDice (ttgf26b, 5 tasks): the cocotb testbench helper cocotb_iverilog_dump was merged into the design

The remaining task IDs keep their original numbering from NotSoTiny-26-07, so the numbering has gaps.

Additional Usage

from datasets import load_dataset

# Load the complete dataset with all shuttles (1,128 tasks)
ds = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", split="test")

# Or load a specific shuttle
ds_ttsky25b = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttsky25b", split="test") # 140 tasks
ds_ttgf0p2  = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttgf0p2",  split="test") # 46 tasks
ds_ttihp26a = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttihp26a", split="test") # 191 tasks
ds_ttsky26a = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttsky26a", split="test") # 233 tasks
ds_ttihp0p4 = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttihp0p4", split="test") # 12 tasks
ds_ttgf26a  = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttgf26a",  split="test") # 84 tasks
ds_ttsky26b = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttsky26b", split="test") # 328 tasks
ds_ttgf0p3  = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttgf0p3",  split="test") # 3 tasks
ds_ttgf26b  = load_dataset("HPAI-BSC/NotSoTiny-26-07-CVDP", "ttgf26b",  split="test") # 91 tasks

License

The dataset is released under the Apache License 2.0.

Citation Information

@misc{ghorab2025notsotinylargelivingbenchmark,
      title={NotSoTiny: A Large, Living Benchmark for RTL Code Generation}, 
      author={Razine Moundir Ghorab and Emanuele Parisi and Cristian Gutierrez-Gomez and Miquel Albert\'i-Binimelis and Miquel Moreto and Dario Garcia-Gasulla and Gokcen Kestor},
      year={2025},
      eprint={2512.20823},
      archivePrefix={arXiv},
      primaryClass={cs.AR},
      url={[https://arxiv.org/abs/2512.20823](https://arxiv.org/abs/2512.20823)}, 
}

Acknowledgements

The HPAI team behind NotSoTiny would like to thank the Tiny Tapeout community for the open source efforts, which made possible this contribution. Special thanks to Matt Venn for his support.

We would also like to thank the CVDP team at NVIDIA for developing and open-sourcing the CVDP framework and to the Si2 Coalition Extend and Expand working group.

Downloads last month
-

Collection including HPAI-BSC/NotSoTiny-26-07-CVDP

Paper for HPAI-BSC/NotSoTiny-26-07-CVDP