datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
swev-trm-trajectories-25models
SWE-Bench Verified TRM Trajectories (25 Models, Verified Labels)
Trajectories from 25 LLMs attempting SWE-Bench Verified tasks, formatted for
training a Trajectory Reward Model (TRM). Each record is one model's full
multi-turn attempt at one task, labeled with the real SWE-bench harness
verdict (scores.resolved).
Splits
Split
Records
Tasks
Pos
Neg
train
10,107
405
6,108
3,999
val
2,366
95
1,440
926
Train/val are task-disjoint (stable hash on task_id… See the full description on the dataset page: https://huggingface.co/datasets/tarsur385/swev-trm-trajectories-25models.dl-trm-phase2-codebook-v32
DL-TRM Phase 2 Codebook V32
This standalone dataset contains the Phase 2 discrete Z traces for V32.
The Hugging Face dataset viewer reads data/train.jsonl.
Each row contains:
raw_puzzle_id
route_id
route_local_id
medoid_old_id
selected_stage_a_example_index
cluster_size
vocab_size
z_trace: a length-16 list of discrete Z token IDs
The original PyTorch artifacts remain in the repo:
z_traces.pt
codebook.pt
transition_vq_model.pt
diagnostics.json
z_trace_manifest.json
checkpoints/
dl-trm-phase2-codebook-v16
DL-TRM Phase 2 Codebook V16
This standalone dataset contains the Phase 2 discrete Z traces for V16.
The Hugging Face dataset viewer reads data/train.jsonl.
Each row contains:
raw_puzzle_id
route_id
route_local_id
medoid_old_id
selected_stage_a_example_index
cluster_size
vocab_size
z_trace: a length-16 list of discrete Z token IDs
The original PyTorch artifacts remain in the repo:
z_traces.pt
codebook.pt
transition_vq_model.pt
diagnostics.json
z_trace_manifest.json
checkpoints/
dl-trm-phase2-codebook-v256
DL-TRM Phase 2 Codebook V256
This standalone dataset contains the Phase 2 discrete Z traces for V256.
The Hugging Face dataset viewer reads data/train.jsonl.
Each row contains:
raw_puzzle_id
route_id
route_local_id
medoid_old_id
selected_stage_a_example_index
cluster_size
vocab_size
z_trace: a length-16 list of discrete Z token IDs
The original PyTorch artifacts remain in the repo:
z_traces.pt
codebook.pt
transition_vq_model.pt
diagnostics.json
z_trace_manifest.json
checkpoints/
TRM-modified-datamix-tokenized
TRM modified datamix (tokenized)
Pre-tokenized reasoning/pretraining mixture for from-scratch TRM (Tiny Recursive Model)
training, built by running data_io — the HRM-Text data
pipeline — verbatim on sapientinc/HRM-Text-data-io-cleaned-20260515, with three
deliberate, documented deviations (below).
It is emitted in the V1 tokenized dataset format (a single concatenated token pool +
per-epoch document indices) and is ready to stream directly into training — no re-tokenization.… See the full description on the dataset page: https://huggingface.co/datasets/m-ric/TRM-modified-datamix-tokenized.dl-trm-phase2-codebook-v128
DL-TRM Phase 2 Codebook V128
This standalone dataset contains the Phase 2 discrete Z traces for V128.
The Hugging Face dataset viewer reads data/train.jsonl.
Each row contains:
raw_puzzle_id
route_id
route_local_id
medoid_old_id
selected_stage_a_example_index
cluster_size
vocab_size
z_trace: a length-16 list of discrete Z token IDs
The original PyTorch artifacts remain in the repo:
z_traces.pt
codebook.pt
transition_vq_model.pt
diagnostics.json
z_trace_manifest.json
checkpoints/
dl-trm-phase2-controller-kmeans-v128
DL-TRM Phase 2 Controller K-Means V128
Discrete Z traces produced by k-means over Phase 1 controller deltas.
Source trace artifact: phase1_real_medoid_route_trace_shards
Input signal: medoid_controller_delta_trace[1:]
Input shape before quantization: [11400, 16, 512]
Output Z trace shape: [11400, 16]
Vocabulary size: 128
Token id base: 0
Diagnostics
{
"used_codes": 128,
"dead_codes": 0,
"perplexity": 123.55767954902056,
"unique_traces": 9222… See the full description on the dataset page: https://huggingface.co/datasets/omrisap/dl-trm-phase2-controller-kmeans-v128.trm-mastering-chessdl-trm-phase2-controller-kmeans-v16
DL-TRM Phase 2 Controller K-Means V16
Discrete Z traces produced by k-means over Phase 1 controller deltas.
Source trace artifact: phase1_real_medoid_route_trace_shards
Input signal: medoid_controller_delta_trace[1:]
Input shape before quantization: [11400, 16, 512]
Output Z trace shape: [11400, 16]
Vocabulary size: 16
Token id base: 0
Diagnostics
{
"used_codes": 16,
"dead_codes": 0,
"perplexity": 15.491387192451063,
"unique_traces": 9176… See the full description on the dataset page: https://huggingface.co/datasets/omrisap/dl-trm-phase2-controller-kmeans-v16.dl-trm-phase2-controller-kmeans-v32
DL-TRM Phase 2 Controller K-Means V32
Discrete Z traces produced by k-means over Phase 1 controller deltas.
Source trace artifact: phase1_real_medoid_route_trace_shards
Input signal: medoid_controller_delta_trace[1:]
Input shape before quantization: [11400, 16, 512]
Output Z trace shape: [11400, 16]
Vocabulary size: 32
Token id base: 0
Diagnostics
{
"used_codes": 32,
"dead_codes": 0,
"perplexity": 31.17811290262774,
"unique_traces": 9191… See the full description on the dataset page: https://huggingface.co/datasets/omrisap/dl-trm-phase2-controller-kmeans-v32.dl-trm-phase2-controller-kmeans-v256
DL-TRM Phase 2 Controller K-Means V256
Discrete Z traces produced by k-means over Phase 1 controller deltas.
Source trace artifact: phase1_real_medoid_route_trace_shards
Input signal: medoid_controller_delta_trace[1:]
Input shape before quantization: [11400, 16, 512]
Output Z trace shape: [11400, 16]
Vocabulary size: 256
Token id base: 0
Diagnostics
{
"used_codes": 256,
"dead_codes": 0,
"perplexity": 244.03568219976052,
"unique_traces": 9225… See the full description on the dataset page: https://huggingface.co/datasets/omrisap/dl-trm-phase2-controller-kmeans-v256.