Jack Hsieh
JackHsieh
·
AI & ML interests
None yet
Recent Activity
updated a collection about 1 hour ago
Prestar (4B, 80M, statML-arxiv) updated a collection about 1 hour ago
Prestar (4B, 80M, statML-arxiv) updated a collection about 1 hour ago
Prestar (4B, 40M, peS2o-humanities)Organizations
Prestar (4B, 80M, statML-arxiv)
-
JackHsieh/4B-general-paragraph-nothink-prefill.k-8.statml-arxiv-iid.qwen3-ids
Viewer • Updated • 44.7M • 605 -
JackHsieh/4B-base.4B-gpnp-kv-tags-thought-thoughts.iid-80M-sans-256.stride-1.suffix-conditional
4B • Updated • 26 -
JackHsieh/Qwen3-4B.gpnp-rl.consumer-4B-base-gpnp-kvtt-iid80M-sans256.heldout256.lr3e-7-kl0.step-1472
Text Generation • 4B • Updated
Prestar (4B, 40M, statML-arxiv)
-
JackHsieh/4B-general-paragraph-nothink-prefill.k-8.statml-arxiv-iid.qwen3-ids
Viewer • Updated • 44.7M • 605 -
JackHsieh/4B-base.4B-gpnp-kv-tags-thought-thoughts.iid-40M-sans-256.stride-1.suffix-conditional
4B • Updated • 44 • 1 -
JackHsieh/Qwen3-4B.gpnp-rl.consumer-4B-base-gpnp-kvtt-iid40M-sans256.lr7e-7-kl0.step-960
Text Generation • 4B • Updated • 818 -
JackHsieh/4B-gpnp-rl-step960.k-8.statml-arxiv-iid.qwen3-ids.kv-tags-thought
Viewer • Updated • 12.4M • 319
k=8 statML-arxiv prepended thoughts
Sets of thoughts used in prestar experiments at k=8 and their kv-explained variants.
Name format: [generator].[stride]
-
JackHsieh/4B-ranked-v9b.stride-train2-test32.k-8.statml-arxiv.qwen3-ids.kv-tags-explained
Viewer • Updated • 5.03M • 175 -
JackHsieh/4B-ranked-v9b.stride-train2-test32.k-8.statml-arxiv.qwen3-ids
Viewer • Updated • 5.03M • 297 -
JackHsieh/4B-Instruct-distill-3rkrz2vo.stride-1.k-8.statml-arxiv.qwen3-ids.kv-tags-explained
Viewer • Updated • 12.4M • 437 -
JackHsieh/4B-Instruct-distill-3rkrz2vo.stride-1.k-8.statml-arxiv.qwen3-ids
Viewer • Updated • 12.4M • 699
debug, k=8 statML-arxiv thoughts
TissueMNIST-vlm-features
Qwen3-4B-Instruct-2507.statml-arxiv.FAKE-thoughts
Qwen3-4B-Instruct-2507.statml-arxiv.thoughts
Datasets of the form Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-[chunk size].L-[thought length].statml-arxiv
-
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-64.L-1024.statml-arxiv
Viewer • Updated • 5.33M • 205 -
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-64.L-256.statml-arxiv
Viewer • Updated • 5.24M • 109 -
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-128.L-1024.statml-arxiv
Viewer • Updated • 2.58M • 171 -
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-128.L-512.statml-arxiv
Viewer • Updated • 2.58M • 118
Prestar (4B, 160M, statML-arxiv)
Prestar (8B, 40M, statML-arxiv)
-
JackHsieh/8B-general-paragraph-nothink-prefill.k-8.statml-arxiv-iid.qwen3-ids.kv-tags-thought
Viewer • Updated • 12.4M • 260 -
JackHsieh/8B-base.8B-gpnp-kv-tags-thought-thoughts.iid-40M-sans-256.stride-1.suffix-conditional
8B • Updated • 25 -
JackHsieh/Qwen3-8B.gpnp-rl.consumer-8B-base-gpnp-kvtt-iid40M-sans256.lr7e-7-kl0.step-1664
Text Generation • 8B • Updated
Prestar finalized
statML-arxiv
Subsets of the statML-arxiv datasplit. Unless otherwise indicated, these are tokenized under Qwen3's tokenizer.
32B, k=8 stat-ML-arxiv thoughts
Thoughts generated from 32B, for chunk size 8, on statML-arxiv-40-20M
-
JackHsieh/32B-predict.rule-r-1.0-k-8.L-16.statml-arxiv
Viewer • Updated • 22.4M • 64 -
JackHsieh/32B-predict-revise.rule-r-0.03125-k-8.L-512.statml-arxiv
Viewer • Updated • 311k • 24 -
JackHsieh/32B-predict-revise-v2.rule-r-1.0-k-8.L-512.statml-arxi
Viewer • Updated • 2.49M • 64 -
JackHsieh/32B-propose-choose.rule-r-0.03125-k-8.L-512.statml-arxiv
Viewer • Updated • 311k • 32
dclm-replay
Qwen3-4B-Instruct-2507.statml-arxiv.thoughts.DEBUG
-
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-256.L-1024.statml-arxiv-debug-128
Viewer • Updated • 11.5k • 20 -
JackHsieh/FAKE-oracle.rule-thoughtful-except-first.k-256.L-1024.statml-arxiv-debug-128-test
Viewer • Updated • 3.84k • 16 -
JackHsieh/FAKE-noise.rule-thoughtful-except-first.k-256.L-1024.statml-arxiv-debug-128-test
Viewer • Updated • 3.84k • 18 -
JackHsieh/FAKE-mantra.rule-thoughtful-except-first.k-256.L-1024.statml-arxiv-debug-128-test
Viewer • Updated • 3.84k • 20
Prestar (4B, 40M, peS2o-humanities)
Prestar (4B, 160M, statML-arxiv)
Prestar (4B, 80M, statML-arxiv)
-
JackHsieh/4B-general-paragraph-nothink-prefill.k-8.statml-arxiv-iid.qwen3-ids
Viewer • Updated • 44.7M • 605 -
JackHsieh/4B-base.4B-gpnp-kv-tags-thought-thoughts.iid-80M-sans-256.stride-1.suffix-conditional
4B • Updated • 26 -
JackHsieh/Qwen3-4B.gpnp-rl.consumer-4B-base-gpnp-kvtt-iid80M-sans256.heldout256.lr3e-7-kl0.step-1472
Text Generation • 4B • Updated
Prestar (8B, 40M, statML-arxiv)
-
JackHsieh/8B-general-paragraph-nothink-prefill.k-8.statml-arxiv-iid.qwen3-ids.kv-tags-thought
Viewer • Updated • 12.4M • 260 -
JackHsieh/8B-base.8B-gpnp-kv-tags-thought-thoughts.iid-40M-sans-256.stride-1.suffix-conditional
8B • Updated • 25 -
JackHsieh/Qwen3-8B.gpnp-rl.consumer-8B-base-gpnp-kvtt-iid40M-sans256.lr7e-7-kl0.step-1664
Text Generation • 8B • Updated
Prestar (4B, 40M, statML-arxiv)
-
JackHsieh/4B-general-paragraph-nothink-prefill.k-8.statml-arxiv-iid.qwen3-ids
Viewer • Updated • 44.7M • 605 -
JackHsieh/4B-base.4B-gpnp-kv-tags-thought-thoughts.iid-40M-sans-256.stride-1.suffix-conditional
4B • Updated • 44 • 1 -
JackHsieh/Qwen3-4B.gpnp-rl.consumer-4B-base-gpnp-kvtt-iid40M-sans256.lr7e-7-kl0.step-960
Text Generation • 4B • Updated • 818 -
JackHsieh/4B-gpnp-rl-step960.k-8.statml-arxiv-iid.qwen3-ids.kv-tags-thought
Viewer • Updated • 12.4M • 319
Prestar finalized
k=8 statML-arxiv prepended thoughts
Sets of thoughts used in prestar experiments at k=8 and their kv-explained variants.
Name format: [generator].[stride]
-
JackHsieh/4B-ranked-v9b.stride-train2-test32.k-8.statml-arxiv.qwen3-ids.kv-tags-explained
Viewer • Updated • 5.03M • 175 -
JackHsieh/4B-ranked-v9b.stride-train2-test32.k-8.statml-arxiv.qwen3-ids
Viewer • Updated • 5.03M • 297 -
JackHsieh/4B-Instruct-distill-3rkrz2vo.stride-1.k-8.statml-arxiv.qwen3-ids.kv-tags-explained
Viewer • Updated • 12.4M • 437 -
JackHsieh/4B-Instruct-distill-3rkrz2vo.stride-1.k-8.statml-arxiv.qwen3-ids
Viewer • Updated • 12.4M • 699
statML-arxiv
Subsets of the statML-arxiv datasplit. Unless otherwise indicated, these are tokenized under Qwen3's tokenizer.
debug, k=8 statML-arxiv thoughts
32B, k=8 stat-ML-arxiv thoughts
Thoughts generated from 32B, for chunk size 8, on statML-arxiv-40-20M
-
JackHsieh/32B-predict.rule-r-1.0-k-8.L-16.statml-arxiv
Viewer • Updated • 22.4M • 64 -
JackHsieh/32B-predict-revise.rule-r-0.03125-k-8.L-512.statml-arxiv
Viewer • Updated • 311k • 24 -
JackHsieh/32B-predict-revise-v2.rule-r-1.0-k-8.L-512.statml-arxi
Viewer • Updated • 2.49M • 64 -
JackHsieh/32B-propose-choose.rule-r-0.03125-k-8.L-512.statml-arxiv
Viewer • Updated • 311k • 32
TissueMNIST-vlm-features
dclm-replay
Qwen3-4B-Instruct-2507.statml-arxiv.FAKE-thoughts
Qwen3-4B-Instruct-2507.statml-arxiv.thoughts.DEBUG
-
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-256.L-1024.statml-arxiv-debug-128
Viewer • Updated • 11.5k • 20 -
JackHsieh/FAKE-oracle.rule-thoughtful-except-first.k-256.L-1024.statml-arxiv-debug-128-test
Viewer • Updated • 3.84k • 16 -
JackHsieh/FAKE-noise.rule-thoughtful-except-first.k-256.L-1024.statml-arxiv-debug-128-test
Viewer • Updated • 3.84k • 18 -
JackHsieh/FAKE-mantra.rule-thoughtful-except-first.k-256.L-1024.statml-arxiv-debug-128-test
Viewer • Updated • 3.84k • 20
Qwen3-4B-Instruct-2507.statml-arxiv.thoughts
Datasets of the form Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-[chunk size].L-[thought length].statml-arxiv
-
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-64.L-1024.statml-arxiv
Viewer • Updated • 5.33M • 205 -
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-64.L-256.statml-arxiv
Viewer • Updated • 5.24M • 109 -
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-128.L-1024.statml-arxiv
Viewer • Updated • 2.58M • 171 -
JackHsieh/Qwen3-4B-Instruct-2507.rule-thoughtful-except-first.k-128.L-512.statml-arxiv
Viewer • Updated • 2.58M • 118