AbstractPhil commited on
Commit
fe4f66d
·
verified ·
1 Parent(s): 68a434c

Burndown record completion: every s1 replication ckpt, every monolith/control falsifier, exp011 fused stacks, exp012/013 blob+uniform stacks, exp014/015 key banks, exp004 fork run saves + LoRA control (salvaged from the pod before window close)

Browse files
Files changed (40) hide show
  1. exp001_sd15_relay/relay_all16_s1.pt +3 -0
  2. exp004_anima_relay/ckpt/aleph_anima_512.toml +50 -0
  3. exp004_anima_relay/ckpt/epoch1/aleph_relay.pt +3 -0
  4. exp004_anima_relay/ckpt/epoch1/aleph_relay.safetensors +3 -0
  5. exp004_anima_relay/ckpt/epoch4/aleph_relay.pt +3 -0
  6. exp004_anima_relay/ckpt/epoch4/aleph_relay.safetensors +3 -0
  7. exp004_anima_relay/ckpt/exp004_dataset.toml +20 -0
  8. exp004_anima_relay/ckpt/step2500/aleph_relay.pt +3 -0
  9. exp004_anima_relay/ckpt/step2500/aleph_relay.safetensors +3 -0
  10. exp004_anima_relay/lora_ctrl/epoch4/adapter_model.safetensors +3 -0
  11. exp004_anima_relay/lora_ctrl/step2500/adapter_model.safetensors +3 -0
  12. exp006_sd15core_relay/lora_r32_s0/adapter_model.safetensors +3 -0
  13. exp006_sd15core_relay/lora_r32_s1/adapter_model.safetensors +3 -0
  14. exp006_sd15core_relay/relay_all16_s1.pt +3 -0
  15. exp007_amoe_lora/amoe_core_s1.pt +3 -0
  16. exp007_amoe_lora/mono_core_s1.pt +3 -0
  17. exp008_multiband/mb3_s1.pt +3 -0
  18. exp008_multiband/mono48_s0.pt +3 -0
  19. exp008_multiband/mono48_s1.pt +3 -0
  20. exp009_bandroles/roles_s1.pt +3 -0
  21. exp011_fused_multiband/monolith_s0.pt +3 -0
  22. exp011_fused_multiband/monolith_s1.pt +3 -0
  23. exp011_fused_multiband/multiband3_s0.pt +3 -0
  24. exp011_fused_multiband/multiband3_s1.pt +3 -0
  25. exp012_blob_supervised/blob_mb3_s0.pt +3 -0
  26. exp012_blob_supervised/blob_mb3_s1.pt +3 -0
  27. exp013_blob_flow/blob_mb3_s0.pt +3 -0
  28. exp013_blob_flow/blob_mb3_s1.pt +3 -0
  29. exp013_blob_flow/uniform_mb3_s0.pt +3 -0
  30. exp013_blob_flow/uniform_mb3_s1.pt +3 -0
  31. exp014_te_dispatch/addr_key_s0.pt +3 -0
  32. exp014_te_dispatch/addr_key_s1.pt +3 -0
  33. exp014_te_dispatch/sigma_key_s0.pt +3 -0
  34. exp014_te_dispatch/sigma_key_s1.pt +3 -0
  35. exp014_te_dispatch/te_key_s0.pt +3 -0
  36. exp014_te_dispatch/te_key_s1.pt +3 -0
  37. exp015_router_v2/mhat_addr_s0.pt +3 -0
  38. exp015_router_v2/mhat_addr_s1.pt +3 -0
  39. exp015_router_v2/state_sigma_s0.pt +3 -0
  40. exp015_router_v2/state_sigma_s1.pt +3 -0
exp001_sd15_relay/relay_all16_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e441b38812efd545443e1a5bec5c7231b5334b096ebe62c13c1e759d12434120
3
+ size 12979593
exp004_anima_relay/ckpt/aleph_anima_512.toml ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # exp004 (r2 diffusion aleph line): Anima 2B DiT + per-block aleph relays.
2
+ # Frozen trunk (every non-relay lr = 0), fp32 relay adapters (zero-init W+b,
3
+ # gate -3.0), pure Adam wd=0 (the 'adam' branch added on feat/aleph-adapter).
4
+ # Anima weights are NC (CircleStone NC + NVIDIA OML) — derived checkpoints too.
5
+
6
+ output_dir = '/workspace/ckpts2/aleph_anima_512'
7
+ dataset = '/workspace/geolip2/pod2/exp004_dataset.toml'
8
+
9
+ epochs = 4
10
+ max_steps = 3000
11
+ micro_batch_size_per_gpu = 4
12
+ pipeline_stages = 1
13
+ gradient_accumulation_steps = 1
14
+ gradient_clipping = 0
15
+ save_every_n_steps = 2500
16
+ checkpoint_every_n_minutes = 60
17
+ activation_checkpointing = true
18
+ save_dtype = 'bfloat16' # dtype law: match the trunk
19
+ caching_batch_size = 8
20
+
21
+ [model]
22
+ type = 'anima'
23
+ # paths per repos/anima-trainer.md (circlestone-labs/Anima split_files/);
24
+ # fill in the pod-local absolute paths at launch:
25
+ transformer_path = '/workspace/models/anima/split_files/diffusion_models/anima-base-v1.0.safetensors'
26
+ llm_path = '/workspace/models/anima/split_files/text_encoders/qwen_3_06b_base.safetensors'
27
+ vae_path = '/workspace/models/anima/split_files/vae/qwen_image_vae.safetensors'
28
+ dtype = 'bfloat16'
29
+ timestep_sample_method = 'logit_normal'
30
+
31
+ # the aleph relay stack (models/aleph_relay.py)
32
+ aleph_relay = true
33
+ aleph_relay_every = 1 # 'every2' is the param-efficiency arm
34
+ aleph_relay_lr = 1e-3
35
+ # freeze the trunk + the llm_adapter (non-negotiable: repos/anima-trainer.md)
36
+ self_attn_lr = 0
37
+ cross_attn_lr = 0
38
+ mlp_lr = 0
39
+ mod_lr = 0
40
+ llm_adapter_lr = 0
41
+ # aleph_relay_path = '...' # resume a saved stack
42
+
43
+ [optimizer]
44
+ type = 'adam' # pure Adam — NEVER adamw on aleph paths
45
+ lr = 0 # base bucket frozen; relays use aleph_relay_lr
46
+ betas = [0.9, 0.99]
47
+ weight_decay = 0.0
48
+
49
+ [monitoring]
50
+ enable_wandb = false
exp004_anima_relay/ckpt/epoch1/aleph_relay.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:08eb75800f8df631a87c2957980c0a7446f326d0a40cd1387a400216c019908e
3
+ size 28622287
exp004_anima_relay/ckpt/epoch1/aleph_relay.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2191af682c63d4f6229f6ab293b0e93e6cb04e947a646bd4120d05ba8a63688a
3
+ size 28577712
exp004_anima_relay/ckpt/epoch4/aleph_relay.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35757f5a9aa925ba7ad4c940e5aece62b6247a5fa397e08d6ea591fe72ed18a0
3
+ size 28622287
exp004_anima_relay/ckpt/epoch4/aleph_relay.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:251566a5652db0f2785476684ff4e39a3d955d6ea5165f9ba2d74a5566b299cf
3
+ size 28577712
exp004_anima_relay/ckpt/exp004_dataset.toml ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # exp004: local-parquet slice of ft1 qwen_90k (2 shards = 3000 rows — the
2
+ # natural row cap; the fork's HF source has no limit key).
3
+ resolutions = [512]
4
+ enable_ar_bucket = true
5
+ min_ar = 0.5
6
+ max_ar = 2.0
7
+ num_ar_buckets = 7
8
+ cache_backend = 'parquet'
9
+ cache_shard_size_mb = 350
10
+
11
+ [[directory]]
12
+ type = 'parquet'
13
+ parquet_files = '/workspace/data/ft1_shards/*.parquet'
14
+ image_column = 'image'
15
+ width_column = 'image_width'
16
+ height_column = 'image_height'
17
+ caption_column = 'caption_vlm_json'
18
+ caption_type = 'text'
19
+ num_repeats = 1
20
+ skip_empty_caption = true
exp004_anima_relay/ckpt/step2500/aleph_relay.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f83ec48a1543ea0fce7f47b2ddbe10da92cf7ac73a1cf886da6e4f9abc2b1941
3
+ size 28622287
exp004_anima_relay/ckpt/step2500/aleph_relay.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7e4d9a94d54ca148b2d1aa3caa1ed3d8ee4b3ecdb2867a1d139205dc30196b3d
3
+ size 28577712
exp004_anima_relay/lora_ctrl/epoch4/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1db102fd0c94aea3feea87ac92e375d6296430570df8011c74e973b9fb8d8a58
3
+ size 74518360
exp004_anima_relay/lora_ctrl/step2500/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e0d7266de5a228a12b2d3b746a02e743117be31eb5870322e1d5be6abe9fc58a
3
+ size 74518360
exp006_sd15core_relay/lora_r32_s0/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57dbc3eacced018ca969f7237b763cd81c9fdd53397c015a9bb0c22c883b801a
3
+ size 12750664
exp006_sd15core_relay/lora_r32_s1/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:97da42238faf35d8a60979695c4ff188346328a50eddbe6f96da9596701df310
3
+ size 12750664
exp006_sd15core_relay/relay_all16_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dfced5e9085dcaa58a95de6fa9fedb2fd35b39ae5d3446d2ae5a6f29e75a489f
3
+ size 12979593
exp007_amoe_lora/amoe_core_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5d57885f46225347035db3c3adc3bab8548e64f9f3ad5d14d06a04a8801bae2d
3
+ size 3681565
exp007_amoe_lora/mono_core_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:427662e3e873ac8eef575ce2e3cfa0890039ec3fb223c103988886226191022f
3
+ size 3265793
exp008_multiband/mb3_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5e7969aaba3f95c4c293b2cd093dd49a0316e0a24c6d255e8b81b67e9a3b4347
3
+ size 4988505
exp008_multiband/mono48_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fae3a04d9537969bc8bb0d1ca7ff74aa16e29c2aafe84e6b764c3db5b66b509c
3
+ size 4863023
exp008_multiband/mono48_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61156e10d03b48a9ce27097a7b08cb31cf86f003d5bd5320a82781dd138feb0f
3
+ size 4863023
exp009_bandroles/roles_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1eea5a3f9b6a43000d652453d5c41584bfa2741320fb6208ba1348efe903b818
3
+ size 4994597
exp011_fused_multiband/monolith_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c5f0fed19e752b96b14b1c666cbc52bf67d293699d5c5230c282b609dc44d60e
3
+ size 4863163
exp011_fused_multiband/monolith_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c88ab36703354aa5255fcc6473e0adf303f2555a8563c3e4d50d2119d9b77c97
3
+ size 4863163
exp011_fused_multiband/multiband3_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a93fd4149f774ead3e5af8b73b0c06d2fa4771d448c19435bd424074c6a11c77
3
+ size 4995427
exp011_fused_multiband/multiband3_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5318ab85434b6919c760f990d7df713d1ae367057bd3f5b2dd6c64b25d3b4fe0
3
+ size 4995427
exp012_blob_supervised/blob_mb3_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8394e41e325da9e04c488112c55925ea1775f10de0d78be5aecfaa6097f1f4f4
3
+ size 4995095
exp012_blob_supervised/blob_mb3_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b5a1eea3d6220da541d36c60848961fc1f3a0976187156be840ac58f2916f2b
3
+ size 4995095
exp013_blob_flow/blob_mb3_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e480a7c8b5e330f71cc9e03bbbd83c65f810f354492a557f2b6887d2d9887245
3
+ size 4995095
exp013_blob_flow/blob_mb3_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fda39788f1cb2a2a801c1242b8259619a0015d6ca9583b5975e1a3bc3adae65c
3
+ size 4995095
exp013_blob_flow/uniform_mb3_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a6bbe5da0fdd8cfcc55e19015a64cb39a7a677c334c9830e3742cc2bdd68b03
3
+ size 4995593
exp013_blob_flow/uniform_mb3_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe236fc283fbc774ffc480a83f8d1b8aad7bb6144d232f55fb7480d572a7fd3a
3
+ size 4995593
exp014_te_dispatch/addr_key_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0e0a96c6b10090fd7afea5fff083c717b932c0f8761ea914e4622df5df222768
3
+ size 4735351
exp014_te_dispatch/addr_key_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6cc097a73130fa91c7293b95f7f9db496e5ee0f0022802896173b81536b1414b
3
+ size 4735351
exp014_te_dispatch/sigma_key_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f6b7ff5976ce5d7ed671ce9dd60283f7da113566f0e4e2704e8e046d9030ef03
3
+ size 3681565
exp014_te_dispatch/sigma_key_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3d7e2c768e3643f264dcf7cc6e49e4f7f4b3b436c6cf47a6772e532cd3844374
3
+ size 3681565
exp014_te_dispatch/te_key_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f602e8d8a75dcf2e646ec28c20442a9dafb00e3c2518e82c7add16d434fa695
3
+ size 3882827
exp014_te_dispatch/te_key_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe19b6b15c822bed648840536cca0bc2458c3fb81c7fac8d401cb02998f80d36
3
+ size 3882827
exp015_router_v2/mhat_addr_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a206e53a3e35543463fab8bbd69bb4699713588e87cf73f5a722ce726cba716
3
+ size 3724157
exp015_router_v2/mhat_addr_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f768f2a055cddfe454ee54023bf49f100d27bc45bb78eb9d0e47caffef44feaf
3
+ size 3724157
exp015_router_v2/state_sigma_s0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:652246decc809d30961205bfbefcb1c678150d497798c3bae9f003a617475735
3
+ size 3682089
exp015_router_v2/state_sigma_s1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ac4311921355124b4d9852086ae1fd665f9bee77465c7be8394e4da0db32acd
3
+ size 3682089