import arm qwen3/frame_k16_solo_o0: anchors (every close), stats, results, log

import/qwen3/frame_k16_solo_o0/closes/qimp_frame_k16_solo_o0_close1_s15_qwen_import.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5becfc18341f29327f184b62b52583f834542b2d7dbd8d54008ada280a299869
+size 54819512

import/qwen3/frame_k16_solo_o0/closes/qimp_frame_k16_solo_o0_close2_s15_qwen_import.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de32a2d1b4acfd94278c1d764855ec3799de952ad7f8bad7cff5c69a82927ba7
+size 54819512

import/qwen3/frame_k16_solo_o0/manifest.json

@@ -0,0 +1,18 @@
+{
+ "tag": "qimp_frame_k16_solo_o0",
+ "run": "frame_k16_solo_o0",
+ "convention": "qwen3",
+ "member": "s15_qwen_import",
+ "final_anchor": "qimp_frame_k16_solo_o0_close2_s15_qwen_import.safetensors",
+ "closes": [
+  "qimp_frame_k16_solo_o0_close1_s15_qwen_import.safetensors",
+  "qimp_frame_k16_solo_o0_close2_s15_qwen_import.safetensors"
+ ],
+ "finished": true,
+ "form": "frame",
+ "base": "solo",
+ "steps": 1117,
+ "trunk_step": 245674,
+ "shipped": "2026-10-08 17:02 local",
+ "note": "every close's anchor ships (mids included); the final close's anchor is the run's s15_qwen_import.safetensors"
+}

import/qwen3/frame_k16_solo_o0/s15_qwen_import.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de32a2d1b4acfd94278c1d764855ec3799de952ad7f8bad7cff5c69a82927ba7
+size 54819512

import/qwen3/frame_k16_solo_o0/stats.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39829da5bbd6a057a2e168ac6db64c08777ff2b1ea363e53c0159d60eec75bc1
+size 264333

import/qwen3/frame_k16_solo_o0/train.log

@@ -0,0 +1,79 @@
+[    0.9s] THE JOB: qimp_frame_k16_solo_o0: the Qwen import arm s15_qwen_import (frame form; the teacher Qwen3 0.6B base at depth 16) alone on the bare trunk; blocks all; 4000 steps, a close every 1000; lambda 2.0 off-domain, 0.0 on the caption rows; pure Adam lr 0.001; the compiled chain with recompute; card 30.5 GB with 30.4 GB free; host RSS 0.6 GB
+Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
+[   17.9s] mini-beatrix-3 @ 245,674: trunk loaded from /workspace/qarm/dl/mini-beatrix-3/checkpoints/step_00245674.safetensors and frozen (376.0M); alephllm 0.10.7 at /workspace/qarm/alephllm/geolip/alephllm; amoe-lora 0.2.11; torch 2.8.0+cu128
+[   18.5s] tokenizer /workspace/qarm/dl/qwen3_tokenizer (gpt2 spelling; vocab 151669; sha256 be75606093db2094): 'a taco truck' -> ['a', 'Ġtaco', 'Ġtruck']; sites [(1, 6, 7)]
+[   28.1s] the teacher: Qwen3 0.6B base (/workspace/qarm/dl/qwen_3_06b_base.safetensors, sha256 cd2a512003e2f9f3; 28 layers, width 1024; config /workspace/qarm/dl/qwen3_06b); depth 16 = the input of layer 16
+[   28.1s] the read's captions: 512 of each draw of /workspace/qarm/dl/coco_caps_2x2048.json (4,090 captions of both draws kept out of training)
+[   40.9s] caption pack (/workspace/qarm/dl/mini-beatrix-3/caption_pack/pack_v2, train): dff 121,153 rows, scf 42,080 rows, danbooru 603,026 rows, cc12m 1,242,176 rows, coco 19,227 rows; host RSS 4.2 GB
+[   45.5s] read rows draw1: 512 captions (0 unspellable or too long), rows (512, 99), 4,920 sites; Qwen's states at ['16', '28post'] (fp32; final-state size 132.0)
+[   49.6s] read rows draw2: 512 captions (0 unspellable or too long), rows (512, 151), 5,086 sites; Qwen's states at ['16', '28post'] (fp32; final-state size 131.5)
+[   49.6s] streams: web text; partners none (two web rows a quiet chunk); held-out sets ['captions', 'web'] x 8 batches
+[arms] attached 's15_qwen_import' for phase 'qwen_import': 13.69M params x 32 blocks, lambda 2.0, spec {'n_slots': 16, 'K': 16, 'D': 8, 'hidden': 256}, recompute on, compile on
+[   70.9s] BIRTH: the member attached over the trunk: web rows read exactly as before; the hand-run trunk against the model's forward: armed 0.0e+00, masked 0.0e+00 (gates: 0.0)
+[   82.3s] standardization statistics from 13,911 reference sites on 8 batches (hers per block; Qwen's depth-16 states: size 61.7, sigma range 0.486..5.375); saved /workspace/qarm/out/anchors/qimp_frame_k16_solo_o0_stats.pt
+[  192.3s] THE FRAME (fp64 ridges Qwen depth 16 -> her standardized states, 5-fold CV over (0.01, 0.1, 1.0, 10.0, 100.0) x trace/p, 110 s): held-out R2 per block 8: 0.381 (lam 0.1), 12: 0.412 (lam 0.1), 16: 0.454 (lam 0.1), 20: 0.485 (lam 0.1), 24: 0.477 (lam 0.1), 28: 0.419 (lam 0.1), 31: 0.327 (lam 0.1); saved /workspace/qarm/out/anchors/qimp_frame_k16_solo_o0_frame.pt
+[  469.8s] CLOSE 0 (step 0): the import read on draw 1 (512 captions, 4,920 sites); per block: her reading against Qwen's final state, masked -> armed / against depth 16 masked -> armed; context (token means removed) masked -> armed
+[  469.8s]    block  8: final 0.426 -> 0.426 / depth 0.496 -> 0.496; context 0.285 -> 0.285; mse 0.749 cos 0.507
+[  469.8s]    block 12: final 0.470 -> 0.470 / depth 0.548 -> 0.548; context 0.331 -> 0.331; mse 0.738 cos 0.542
+[  469.8s]    block 16: final 0.508 -> 0.508 / depth 0.579 -> 0.579; context 0.372 -> 0.372; mse 0.692 cos 0.585
+[  469.8s]    block 18: final 0.514 -> 0.514 / depth 0.581 -> 0.581; context 0.386 -> 0.386; mse 0.684 cos 0.596
+[  469.8s]    block 20: final 0.526 -> 0.526 / depth 0.573 -> 0.573; context 0.399 -> 0.399; mse 0.690 cos 0.601
+[  469.8s]    block 24: final 0.562 -> 0.562 / depth 0.526 -> 0.526; context 0.466 -> 0.466; mse 0.777 cos 0.565
+[  469.8s]    block 28: final 0.540 -> 0.540 / depth 0.452 -> 0.452; context 0.448 -> 0.448; mse 0.798 cos 0.518
+[  469.8s]    block 31: final 0.475 -> 0.475 / depth 0.385 -> 0.385; context 0.385 -> 0.385; mse 0.731 cos 0.475
+[  469.8s]    draw 2 (the repeat), armed against the final state at 12/16/20/24/28: 0.471 / 0.508 / 0.529 / 0.567 / 0.546
+[  469.8s]    I1 the reach: the largest rise over her bare reading +0.000 at block 31 (draw 2 +0.000 at 31) against the bar +.10; I2 the context: +0.000 at block 2 against +.10; bits: captions +0.00000, web +0.00000 against +.012
+[  469.8s]    gate mean 0.047 / widest 0.047; compiled chain: 2 cache entries (limit 256), 0 limit warnings; docs drawn 256 (skipped {'test': 2, 'spelling': 0, 'empty': 0, 'short': 0}); card peak 7.0 GB; host RSS 9.3 GB
+[  469.8s] THE INSTRUMENT CHECK: her bare reading against Qwen's final state on draw 1 (this driver / e038's row): block 8 0.426 / 0.426, block 12 0.470 / 0.470, block 16 0.508 / 0.509, block 20 0.526 / 0.526, block 24 0.562 / 0.562, block 28 0.540 / 0.540; largest difference 0.001
+[stream] opening 'fineweb-edu' (HuggingFaceFW/fineweb-edu) …
+[  476.9s] step 1/4000: task 0.6349 (mse 0.635 cos 0.609); quiet A 0.00000 off 0.00000; 1,727 sites; 7.05 s/step; the job: 7.9 min spent, about 470.2 min of steps left; card peak 16.0 GB
+[  565.1s] step 50/4000: task 0.6324 (mse 0.632 cos 0.609); quiet A 0.00000 off 0.00023; 1,733 sites; 1.90 s/step; the job: 9.4 min spent, about 125.4 min of steps left; card peak 16.1 GB
+[  654.5s] step 100/4000: task 0.6269 (mse 0.627 cos 0.609); quiet A 0.00000 off 0.00042; 1,740 sites; 1.85 s/step; the job: 10.9 min spent, about 120.0 min of steps left; card peak 16.1 GB
+[  744.8s] step 150/4000: task 0.6155 (mse 0.615 cos 0.613); quiet A 0.00000 off 0.00084; 1,738 sites; 1.83 s/step; the job: 12.4 min spent, about 117.6 min of steps left; card peak 16.1 GB
+[  835.1s] step 200/4000: task 0.6089 (mse 0.609 cos 0.612); quiet A 0.00000 off 0.00079; 1,717 sites; 1.83 s/step; the job: 13.9 min spent, about 115.7 min of steps left; card peak 16.1 GB
+[  924.6s] step 250/4000: task 0.5898 (mse 0.590 cos 0.617); quiet A 0.00000 off 0.00073; 1,721 sites; 1.82 s/step; the job: 15.4 min spent, about 113.7 min of steps left; card peak 16.1 GB
+[ 1015.0s] step 300/4000: task 0.5897 (mse 0.590 cos 0.609); quiet A 0.00000 off 0.00091; 1,750 sites; 1.82 s/step; the job: 16.9 min spent, about 112.1 min of steps left; card peak 16.1 GB
+[ 1105.4s] step 350/4000: task 0.5737 (mse 0.574 cos 0.610); quiet A 0.00000 off 0.00095; 1,725 sites; 1.82 s/step; the job: 18.4 min spent, about 110.5 min of steps left; card peak 16.1 GB
+[ 1195.9s] step 400/4000: task 0.5538 (mse 0.554 cos 0.615); quiet A 0.00000 off 0.00085; 1,772 sites; 1.82 s/step; the job: 19.9 min spent, about 108.9 min of steps left; card peak 16.1 GB
+[ 1286.3s] step 450/4000: task 0.5484 (mse 0.548 cos 0.611); quiet A 0.00000 off 0.00092; 1,827 sites; 1.81 s/step; the job: 21.4 min spent, about 107.3 min of steps left; card peak 16.1 GB
+[ 1376.3s] step 500/4000: task 0.5359 (mse 0.536 cos 0.613); quiet A 0.00000 off 0.00085; 1,767 sites; 1.81 s/step; the job: 22.9 min spent, about 105.8 min of steps left; card peak 16.1 GB
+[ 1466.7s] step 550/4000: task 0.5572 (mse 0.557 cos 0.609); quiet A 0.00000 off 0.00093; 1,682 sites; 1.81 s/step; the job: 24.4 min spent, about 104.2 min of steps left; card peak 16.1 GB
+[ 1556.7s] step 600/4000: task 0.5330 (mse 0.533 cos 0.612); quiet A 0.00000 off 0.00097; 1,717 sites; 1.81 s/step; the job: 25.9 min spent, about 102.6 min of steps left; card peak 16.1 GB
+[ 1647.0s] step 650/4000: task 0.5183 (mse 0.518 cos 0.608); quiet A 0.00000 off 0.00095; 1,805 sites; 1.81 s/step; the job: 27.5 min spent, about 101.1 min of steps left; card peak 16.1 GB
+[ 1737.7s] step 700/4000: task 0.5172 (mse 0.517 cos 0.612); quiet A 0.00000 off 0.00096; 1,672 sites; 1.81 s/step; the job: 29.0 min spent, about 99.6 min of steps left; card peak 16.1 GB
+[ 1827.6s] step 750/4000: task 0.5260 (mse 0.526 cos 0.609); quiet A 0.00000 off 0.00096; 1,708 sites; 1.81 s/step; the job: 30.5 min spent, about 98.1 min of steps left; card peak 16.1 GB
+[ 1918.0s] step 800/4000: task 0.4892 (mse 0.489 cos 0.614); quiet A 0.00000 off 0.00097; 1,771 sites; 1.81 s/step; the job: 32.0 min spent, about 96.5 min of steps left; card peak 16.1 GB
+[ 2007.7s] step 850/4000: task 0.4858 (mse 0.486 cos 0.612); quiet A 0.00000 off 0.00098; 1,665 sites; 1.81 s/step; the job: 33.5 min spent, about 95.0 min of steps left; card peak 16.1 GB
+[ 2097.6s] step 900/4000: task 0.4831 (mse 0.483 cos 0.612); quiet A 0.00000 off 0.00106; 1,848 sites; 1.81 s/step; the job: 35.0 min spent, about 93.4 min of steps left; card peak 16.1 GB
+[ 2187.7s] step 950/4000: task 0.4902 (mse 0.490 cos 0.614); quiet A 0.00000 off 0.00112; 1,781 sites; 1.81 s/step; the job: 36.5 min spent, about 91.9 min of steps left; card peak 16.1 GB
+[ 2277.7s] step 1000/4000: task 0.4754 (mse 0.475 cos 0.612); quiet A 0.00000 off 0.00108; 1,745 sites; 1.81 s/step; the job: 38.0 min spent, about 90.4 min of steps left; card peak 16.1 GB
+[ 2277.9s] anchor saved: /workspace/qarm/out/anchors/qimp_frame_k16_solo_o0_close1_s15_qwen_import.safetensors
+[ 2419.4s] CLOSE 1 (step 1000): the import read on draw 1 (512 captions, 4,920 sites); per block: her reading against Qwen's final state, masked -> armed / against depth 16 masked -> armed; context (token means removed) masked -> armed
+[ 2419.4s]    block  8: final 0.426 -> 0.420 / depth 0.496 -> 0.493; context 0.285 -> 0.273; mse 0.551 cos 0.499
+[ 2419.4s]    block 12: final 0.470 -> 0.458 / depth 0.548 -> 0.547; context 0.331 -> 0.309; mse 0.571 cos 0.523
+[ 2419.4s]    block 16: final 0.508 -> 0.500 / depth 0.579 -> 0.585; context 0.372 -> 0.356; mse 0.564 cos 0.585
+[ 2419.4s]    block 18: final 0.514 -> 0.505 / depth 0.581 -> 0.586; context 0.386 -> 0.370; mse 0.562 cos 0.598
+[ 2419.4s]    block 20: final 0.526 -> 0.516 / depth 0.573 -> 0.580; context 0.399 -> 0.387; mse 0.564 cos 0.610
+[ 2419.4s]    block 24: final 0.562 -> 0.550 / depth 0.526 -> 0.544; context 0.466 -> 0.450; mse 0.578 cos 0.590
+[ 2419.4s]    block 28: final 0.540 -> 0.539 / depth 0.452 -> 0.491; context 0.448 -> 0.446; mse 0.511 cos 0.558
+[ 2419.4s]    block 31: final 0.475 -> 0.519 / depth 0.385 -> 0.471; context 0.385 -> 0.427; mse 0.466 cos 0.516
+[ 2419.4s]    draw 2 (the repeat), armed against the final state at 12/16/20/24/28: 0.461 / 0.502 / 0.518 / 0.556 / 0.547
+[ 2419.4s]    I1 the reach: the largest rise over her bare reading +0.043 at block 31 (draw 2 +0.052 at 31) against the bar +.10; I2 the context: +0.042 at block 31 against +.10; bits: captions +0.58302, web +0.00054 against +.012
+[ 2419.4s]    gate mean 0.159 / widest 0.206; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 32,523 (skipped {'test': 44, 'spelling': 0, 'empty': 0, 'short': 0}); card peak 16.1 GB; host RSS 11.4 GB
+[ 2509.2s] step 1050/4000: task 0.4627 (mse 0.463 cos 0.619); quiet A 0.00000 off 0.00109; 1,763 sites; 1.80 s/step; the job: 41.8 min spent, about 88.9 min of steps left; card peak 16.1 GB
+[ 2599.8s] step 1100/4000: task 0.4565 (mse 0.457 cos 0.623); quiet A 0.00000 off 0.00139; 1,731 sites; 1.80 s/step; the job: 43.3 min spent, about 87.4 min of steps left; card peak 16.1 GB
+[ 2630.4s] anchor saved: /workspace/qarm/out/anchors/qimp_frame_k16_solo_o0_close2_s15_qwen_import.safetensors
+[ 2772.1s] CLOSE 2 (step 1117): the import read on draw 1 (512 captions, 4,920 sites); per block: her reading against Qwen's final state, masked -> armed / against depth 16 masked -> armed; context (token means removed) masked -> armed
+[ 2772.1s]    block  8: final 0.426 -> 0.427 / depth 0.496 -> 0.496; context 0.285 -> 0.284; mse 0.612 cos 0.526
+[ 2772.1s]    block 12: final 0.470 -> 0.470 / depth 0.548 -> 0.552; context 0.331 -> 0.326; mse 0.610 cos 0.561
+[ 2772.1s]    block 16: final 0.508 -> 0.513 / depth 0.579 -> 0.588; context 0.372 -> 0.372; mse 0.596 cos 0.609
+[ 2772.1s]    block 18: final 0.514 -> 0.517 / depth 0.581 -> 0.590; context 0.386 -> 0.385; mse 0.593 cos 0.617
+[ 2772.1s]    block 20: final 0.526 -> 0.530 / depth 0.573 -> 0.583; context 0.399 -> 0.401; mse 0.596 cos 0.624
+[ 2772.1s]    block 24: final 0.562 -> 0.570 / depth 0.526 -> 0.541; context 0.466 -> 0.473; mse 0.630 cos 0.598
+[ 2772.1s]    block 28: final 0.540 -> 0.556 / depth 0.452 -> 0.475; context 0.448 -> 0.460; mse 0.581 cos 0.564
+[ 2772.1s]    block 31: final 0.475 -> 0.518 / depth 0.385 -> 0.434; context 0.385 -> 0.420; mse 0.518 cos 0.525
+[ 2772.1s]    draw 2 (the repeat), armed against the final state at 12/16/20/24/28: 0.470 / 0.512 / 0.533 / 0.575 / 0.561
+[ 2772.1s]    I1 the reach: the largest rise over her bare reading +0.042 at block 31 (draw 2 +0.039 at 31) against the bar +.10; I2 the context: +0.035 at block 31 against +.10; bits: captions +0.43317, web +0.00165 against +.012
+[ 2772.1s]    gate mean 0.176 / widest 0.263; compiled chain: 5 cache entries (limit 256), 0 limit warnings; docs drawn 36,225 (skipped {'test': 49, 'spelling': 0, 'empty': 0, 'short': 0}); card peak 16.1 GB; host RSS 11.3 GB
+[ 2772.2s] the stop file /workspace/qarm/STOP_qimp_frame_k16_solo_o0 is present: ending after step 1117
+[ 2772.2s] DONE qimp_frame_k16_solo_o0: 1117 steps in 33.6 min of steps (1.81 s/step); 2 closes; 2772 s in all; card peak 16.1 GB; host RSS 11.3 GB; compiled chain: 5 cache entries (limit 256), 0 limit warnings