WillXH commited on
Commit
bc261a7
·
verified ·
1 Parent(s): 94e7ea4

Upload folder using huggingface_hub

Browse files
Files changed (3) hide show
  1. README.md +10 -5
  2. pusht_2rf_k32_seed0.pt +3 -0
  3. pusht_fm_unet_seed2.pt +3 -0
README.md CHANGED
@@ -8,7 +8,7 @@ tags:
8
  - flow-matching
9
  - push-t
10
  datasets:
11
- - WilliamHangXu/pushf-demos
12
  ---
13
 
14
  # FlowPush checkpoints
@@ -22,16 +22,21 @@ it on the 50 test episodes reproduces that seed's number (the README reports 3-s
22
  | file | model | success@0.95, seed 0 |
23
  |---|---|---|
24
  | `pusht_ddpm_unet_seed0.pt` | DDPM U-Net, state obs, 100 steps | 0.44 |
25
- | `pusht_fm_unet_seed0.pt` | flow-matching U-Net, state obs, 8 Euler steps (the teacher) | 0.60 |
 
26
  | `pusht_fm_dit_seed0.pt` | flow-matching DiT (7.8M params), state obs, 8 steps | 0.34 |
27
  | `pusht_fm_unet_image_seed0.pt` | flow-matching U-Net, 96×96 RGB obs, 8 steps | 0.30 |
28
- | `pusht_2rf_seed0.pt` | 2-rectified flow, 1 step | 0.46 |
29
- | `pusht_1step_distilled_seed0.pt` | 1-step distilled student | 0.52 |
 
30
  | `pushf_scratch_n100_seed0.pt` | Push-F, trained from scratch on 100 demos, 8 steps | 0.66 |
31
  | `pushf_finetune_n100_seed0.pt` | Push-F, fine-tuned from the Push-T teacher on 100 demos (lr 1e-4), 8 steps | 0.78 |
32
 
33
  `eval_records.tgz` holds the per-episode evaluation records behind the repository's
34
- `results/*.csv`.
 
 
 
35
 
36
  ## Usage
37
 
 
8
  - flow-matching
9
  - push-t
10
  datasets:
11
+ - WillXH/pushf-demos
12
  ---
13
 
14
  # FlowPush checkpoints
 
22
  | file | model | success@0.95, seed 0 |
23
  |---|---|---|
24
  | `pusht_ddpm_unet_seed0.pt` | DDPM U-Net, state obs, 100 steps | 0.44 |
25
+ | `pusht_fm_unet_seed0.pt` | flow-matching U-Net, state obs, 8 Euler steps | 0.60 |
26
+ | `pusht_fm_unet_seed2.pt` | flow-matching U-Net, state obs — the distillation teacher | 0.60 |
27
  | `pusht_fm_dit_seed0.pt` | flow-matching DiT (7.8M params), state obs, 8 steps | 0.34 |
28
  | `pusht_fm_unet_image_seed0.pt` | flow-matching U-Net, 96×96 RGB obs, 8 steps | 0.30 |
29
+ | `pusht_2rf_seed0.pt` | 2-rectified flow (K = 8), 1 step | 0.46 |
30
+ | `pusht_2rf_k32_seed0.pt` | 2-rectified flow (K = 32) — parent of the 1-step students | 0.36 |
31
+ | `pusht_1step_distilled_seed0.pt` | 1-step distilled student (trained from `pusht_2rf_k32_seed0.pt`) | 0.52 |
32
  | `pushf_scratch_n100_seed0.pt` | Push-F, trained from scratch on 100 demos, 8 steps | 0.66 |
33
  | `pushf_finetune_n100_seed0.pt` | Push-F, fine-tuned from the Push-T teacher on 100 demos (lr 1e-4), 8 steps | 0.78 |
34
 
35
  `eval_records.tgz` holds the per-episode evaluation records behind the repository's
36
+ `results/*.csv`. All three student seeds share the single K = 32 parent above, which was
37
+ trained from the seed-2 teacher; the repository's `results/lineage.csv` records parents and
38
+ SHA-256 hashes. The Push-T test seeds were reused while developing the distillation recipe,
39
+ so Part 2 comparisons are development-set results.
40
 
41
  ## Usage
42
 
pusht_2rf_k32_seed0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9104d164750fca82ef67d27052003def1cf6474582d5f7275e4ddcd81315f89f
3
+ size 261594999
pusht_fm_unet_seed2.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e79e17a667cc63ffbafbae6d2e4f8b9875eef8e6d86ef9d3a8fb74c9f755ebce
3
+ size 261594871