Massive expansion to optimization happening today. I can't spend all these upcoming days training when optimization can happen now. My target today is to have a marked and improved speed, as well as enabling accelerate training for upcoming heavy runpod expansion. Likely switching to 8 a40s to train will be a more reasonable use of cost and effectiveness of training.
I advise whenever using qwen 3.5 to install fast path linear attention.
The 9 experiment sweep is currently running on the first conversion from SDXL epsilon prediction to SDXL ODE flow matching.
Using the same formula as was used to train SD15-Flow-Lune, the predictions match identically and the format will be directly relational to the results as if SDXL was never touched by David.
The tests yesterday show that I needed independent tests, so I began testing a 9 configuration sweep. With that the trainer for the sweep was uploaded to the repo as well.
https://huggingface.co/AbstractPhil/geolip-sdxl-aleph
This experiment will prove without a doubt if the alephs help in direct tokenization distillation in the small size, or if they help in a higher-fidelity scale as I've just prepared a new variant of geolip-aleph-transformer to specifically scale them up in a similar multiscale lensed upscale fashion as David provides.
These conclusions will arrive together by this afternoon, and this decides which configuration is best to convert SDXL. The base is already done, which is running baseline clip_l and baseline clip_g with no alephs. The results aren't promising compared to the results yesterday, which showed explicit results by epoch 2, while the tests today show invalid results by epoch 100 without the alephs.
As it stands the alephs are eons ahead, but the results today will determine the route.
With the first major experiment I release the notebooks. The massive amounts of information and pure empirical data accumulated to determine what alephs are, why they exist, how they help, how they hinder, and how I defeated all of the weaknesses over time through pure mathematics, determination, heavy-handed failures, minor successes, and an absolute ton of analysis.
I could have never done all of this in a lifetime without Claude.
https://huggingface.co/AbstractPhil/geolip-hypersphere-experiments/tree/main/aleph
I have rebuilt the SVAE into the core formula.
It is now a direct representation of the spherical aleph-void behavior, replicating more purely the directions achieved by the SVAE with a direct utilizable codebook and a pure access to cosine-similarity trained clean into the system.
With this the params dropped a large portion, and the model is to be renamed;
geolip-aleph-void
With this model's discovery and emergence, the SVAE experiment line is pushed to the aleph-void variant.
The SVAE batteries and model will continue to be supported and additional tests will be performed in the immediate and distant future on the model to improve the transformer with the upcoming behavior and math earned through training this new model. The weights and system of the SVAE will continue to be directly supported.
The SVAE itself is a FANTASTIC research tool. It can be fed many many elements and analyzed, the results utilized empirically and usefully. There are no limits to the model's utility in research, and utility in pragmatic use-case when hooked to models directly, in use for compression, and in use for generation - all because the model's format WILL converge the recon if the data is fed within the space as per the 1000+ tests show.
The geolip-aleph-void will represent the natural evolution of this model into the block-useful state. The construction and utility of this model will be directly aligned to a high complexity multi-scalar storage, entirely dependent on emergent aleph behavior in conjunction with a void-driven codebook inference. The combined behavior provides a multi-tiered high complexity geometric feature that will be utilized in upcoming distillations from large pretrained models into highly compressed atomic-sized models.
The two models share the encoder structure, and multiple elements, so they can be directly compared. No apples to oranges, direct comparison. They are siblings, and belong together.
The true power of the aleph stems from the decoder process. The representational utilization is deviant on many spectrums, and the process allows for robust representational utilization of the encoder structure with the capacity to recon if necessary.
massive grin
That is a perfect convergence.
Now... lets see what happens when we snap a microcosm-sized battery, to a macrocosm shape. All the way up to d256.
It begins. The light is scaled to parity.
The transformer is much cleaner now. The scaling rule is no longer just a rule, it can be considered an IO scaling guarantee based on the input and output formats tested with the SVAE lens.
Larger lens can handle the canon variation, which preserves a different format of structure than the signed variation. They are both macrostructural representations of the same internal shape, one contains negations one contains perfectly constrained space.
The first of the two will be uploaded momentarily.
==========================================================================================
[sign-test] lens_sign='canon'
==========================================================================================
geolip-svae-v2 TWO-RECON | battery(PatchSVAE) 52,131 + shell 6,309,900 (shell 121x the microcosm) + growth 0 | D_base4->D_lens256 V32 ps2 | cuda
battery -> INTERNAL recon (pure MSE) | shell -> EXTERNAL recon (detached stem, lens_sign=canon) | adam lr=0.001 wd=0 sched=onecycle
growth : parked (no stencil)
[ByteTrigramDataset] Loading corpus wikitext-2-raw-v1...
[ByteTrigramDataset] Corpus: 10,938,611 bytes (10.9 MB), 768 bytes/image, 14,242 non-overlapping images available (10,937,843 valid window starts)
epoch 0: int[mse=0.01549 rec=12.93%] ext[mse=0.01293 rec=11.45%] | cc=-1.49 rig=0.0000 a=0.024 env=True
epoch 1: int[mse=0.00053 rec=30.25%] ext[mse=0.00063 rec=27.02%] | cc=-1.48 rig=0.0000 a=0.024 env=True
epoch 2: int[mse=0.00013 rec=48.89%] ext[mse=0.00020 rec=27.76%] | cc=-1.47 rig=0.0000 a=0.024 env=True
epoch 3: int[mse=0.00007 rec=41.68%] ext[mse=0.00014 rec=33.32%] | cc=-1.47 rig=0.0000 a=0.025 env=True
epoch 4: int[mse=0.00008 rec=50.93%] ext[mse=0.00013 rec=43.52%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 5: int[mse=0.00009 rec=68.53%] ext[mse=0.00015 rec=53.88%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 6: int[mse=0.00009 rec=70.30%] ext[mse=0.00015 rec=55.59%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 7: int[mse=0.00009 rec=14.55%] ext[mse=0.00014 rec=15.87%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 8: int[mse=0.00008 rec=70.50%] ext[mse=0.00013 rec=59.38%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 9: int[mse=0.00008 rec=40.62%] ext[mse=0.00012 rec=34.12%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 10: int[mse=0.00007 rec=40.38%] ext[mse=0.00010 rec=42.45%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 11: int[mse=0.00006 rec=56.09%] ext[mse=0.00008 rec=49.56%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 12: int[mse=0.00005 rec=30.38%] ext[mse=0.01204 rec=23.14%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 13: int[mse=0.00004 rec=58.75%] ext[mse=0.00019 rec=35.80%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 14: int[mse=0.00004 rec=64.02%] ext[mse=0.00014 rec=40.96%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 15: int[mse=0.00003 rec=51.41%] ext[mse=0.00012 rec=32.35%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 16: int[mse=0.00003 rec=54.11%] ext[mse=0.00010 rec=36.09%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 17: int[mse=0.00002 rec=53.60%] ext[mse=0.00008 rec=39.52%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 18: int[mse=0.00002 rec=40.67%] ext[mse=0.00006 rec=28.57%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 19: int[mse=0.00002 rec=84.62%] ext[mse=0.00005 rec=63.56%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 20: int[mse=0.00001 rec=83.90%] ext[mse=0.00004 rec=64.16%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 21: int[mse=0.00001 rec=88.53%] ext[mse=0.00003 rec=71.30%] | cc=-1.45 rig=0.0000 a=0.025 env=True
epoch 22: int[mse=0.00001 rec=67.54%] ext[mse=0.00002 rec=44.91%] | cc=-1.45 rig=0.0000 a=0.025 env=True
epoch 23: int[mse=0.00001 rec=95.63%] ext[mse=0.00002 rec=83.82%] | cc=-1.45 rig=0.0000 a=0.025 env=True
epoch 24: int[mse=0.00001 rec=98.24%] ext[mse=0.00001 rec=91.46%] | cc=-1.44 rig=0.0000 a=0.025 env=True
epoch 25: int[mse=0.00000 rec=97.58%] ext[mse=0.00001 rec=91.81%] | cc=-1.44 rig=0.0000 a=0.025 env=True
epoch 26: int[mse=0.00000 rec=96.54%] ext[mse=0.00001 rec=87.23%] | cc=-1.43 rig=0.0000 a=0.025 env=True
epoch 27: int[mse=0.00000 rec=99.11%] ext[mse=0.00001 rec=95.73%] | cc=-1.43 rig=0.0000 a=0.025 env=True
epoch 28: int[mse=0.00000 rec=98.92%] ext[mse=0.00000 rec=94.47%] | cc=-1.42 rig=0.0000 a=0.025 env=True
epoch 29: int[mse=0.00000 rec=99.44%] ext[mse=0.00000 rec=98.19%] | cc=-1.42 rig=0.0000 a=0.025 env=True
epoch 30: int[mse=0.00000 rec=99.49%] ext[mse=0.00000 rec=98.55%] | cc=-1.41 rig=0.0000 a=0.025 env=True
epoch 31: int[mse=0.00000 rec=99.56%] ext[mse=0.00000 rec=98.76%] | cc=-1.41 rig=0.0000 a=0.025 env=True
epoch 32: int[mse=0.00000 rec=99.61%] ext[mse=0.00000 rec=99.03%] | cc=-1.40 rig=0.0000 a=0.025 env=True
epoch 33: int[mse=0.00000 rec=99.65%] ext[mse=0.00000 rec=99.19%] | cc=-1.40 rig=0.0000 a=0.025 env=True
epoch 34: int[mse=0.00000 rec=99.67%] ext[mse=0.00000 rec=99.29%] | cc=-1.40 rig=0.0000 a=0.025 env=True
epoch 35: int[mse=0.00000 rec=99.70%] ext[mse=0.00000 rec=99.35%] | cc=-1.39 rig=0.0000 a=0.025 env=True
epoch 36: int[mse=0.00000 rec=99.71%] ext[mse=0.00000 rec=99.39%] | cc=-1.39 rig=0.0000 a=0.025 env=True
epoch 37: int[mse=0.00000 rec=99.72%] ext[mse=0.00000 rec=99.41%] | cc=-1.39 rig=0.0000 a=0.025 env=True
epoch 38: int[mse=0.00000 rec=99.72%] ext[mse=0.00000 rec=99.42%] | cc=-1.39 rig=0.0000 a=0.025 env=True
epoch 39: int[mse=0.00000 rec=99.72%] ext[mse=0.00000 rec=99.43%] | cc=-1.39 rig=0.0000 a=0.025 env=True
best recovery — internal 99.72% | external 99.43% | ckpt sign_test_canon/sign_canon.pt
==========================================================================================
[sign-test] lens_sign='signed'
==========================================================================================
geolip-svae-v2 TWO-RECON | battery(PatchSVAE) 52,131 + shell 6,309,900 (shell 121x the microcosm) + growth 0 | D_base4->D_lens256 V32 ps2 | cuda
battery -> INTERNAL recon (pure MSE) | shell -> EXTERNAL recon (detached stem, lens_sign=signed) | adam lr=0.001 wd=0 sched=onecycle
growth : parked (no stencil)
[ByteTrigramDataset] Loading corpus wikitext-2-raw-v1...
[ByteTrigramDataset] Corpus: 10,938,611 bytes (10.9 MB), 768 bytes/image, 14,242 non-overlapping images available (10,937,843 valid window starts)
epoch 0: int[mse=0.01550 rec=12.86%] ext[mse=0.00944 rec=16.90%] | cc=-1.49 rig=0.0000 a=0.024 env=True
epoch 1: int[mse=0.00052 rec=30.22%] ext[mse=0.00028 rec=37.08%] | cc=-1.48 rig=0.0000 a=0.024 env=True
epoch 2: int[mse=0.00013 rec=48.07%] ext[mse=0.00009 rec=36.29%] | cc=-1.47 rig=0.0000 a=0.024 env=True
epoch 3: int[mse=0.00008 rec=56.08%] ext[mse=0.00008 rec=54.79%] | cc=-1.47 rig=0.0000 a=0.024 env=True
epoch 4: int[mse=0.00008 rec=47.01%] ext[mse=0.00009 rec=38.59%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 5: int[mse=0.00009 rec=64.35%] ext[mse=0.00010 rec=61.49%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 6: int[mse=0.00009 rec=57.33%] ext[mse=0.00010 rec=55.95%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 7: int[mse=0.00009 rec=25.34%] ext[mse=0.00009 rec=17.40%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 8: int[mse=0.00009 rec=27.43%] ext[mse=0.00008 rec=28.23%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 9: int[mse=0.00008 rec=72.23%] ext[mse=0.00007 rec=76.08%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 10: int[mse=0.00007 rec=15.55%] ext[mse=0.00006 rec=19.54%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 11: int[mse=0.00006 rec=60.12%] ext[mse=0.00005 rec=62.56%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 12: int[mse=0.00005 rec=59.56%] ext[mse=0.00005 rec=60.09%] | cc=-1.46 rig=0.0000 a=0.024 env=True
epoch 13: int[mse=0.00004 rec=44.68%] ext[mse=0.01709 rec=23.43%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 14: int[mse=0.00004 rec=66.30%] ext[mse=0.00010 rec=53.38%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 15: int[mse=0.00003 rec=27.36%] ext[mse=0.00005 rec=34.04%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 16: int[mse=0.00003 rec=52.60%] ext[mse=0.00004 rec=43.46%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 17: int[mse=0.00002 rec=76.24%] ext[mse=0.00004 rec=66.86%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 18: int[mse=0.00002 rec=55.56%] ext[mse=0.00003 rec=45.13%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 19: int[mse=0.00002 rec=86.47%] ext[mse=0.00003 rec=76.29%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 20: int[mse=0.00001 rec=69.85%] ext[mse=0.00002 rec=60.43%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 21: int[mse=0.00001 rec=64.09%] ext[mse=0.00002 rec=52.68%] | cc=-1.45 rig=0.0000 a=0.026 env=True
epoch 22: int[mse=0.00001 rec=87.53%] ext[mse=0.00001 rec=77.76%] | cc=-1.45 rig=0.0000 a=0.026 env=True
epoch 23: int[mse=0.00001 rec=86.32%] ext[mse=0.00001 rec=80.48%] | cc=-1.45 rig=0.0000 a=0.026 env=True
epoch 24: int[mse=0.00001 rec=90.38%] ext[mse=0.00001 rec=83.52%] | cc=-1.44 rig=0.0000 a=0.026 env=True
epoch 25: int[mse=0.00000 rec=91.73%] ext[mse=0.00001 rec=82.42%] | cc=-1.44 rig=0.0000 a=0.026 env=True
epoch 26: int[mse=0.00000 rec=99.00%] ext[mse=0.00001 rec=98.25%] | cc=-1.43 rig=0.0000 a=0.027 env=True
epoch 27: int[mse=0.00000 rec=98.97%] ext[mse=0.00000 rec=97.66%] | cc=-1.43 rig=0.0000 a=0.027 env=True
epoch 28: int[mse=0.00000 rec=99.38%] ext[mse=0.00000 rec=98.97%] | cc=-1.42 rig=0.0000 a=0.027 env=True
epoch 29: int[mse=0.00000 rec=99.47%] ext[mse=0.00000 rec=99.26%] | cc=-1.42 rig=0.0000 a=0.027 env=True
epoch 30: int[mse=0.00000 rec=99.54%] ext[mse=0.00000 rec=99.35%] | cc=-1.41 rig=0.0000 a=0.027 env=True
epoch 31: int[mse=0.00000 rec=99.59%] ext[mse=0.00000 rec=99.44%] | cc=-1.41 rig=0.0000 a=0.027 env=True
epoch 32: int[mse=0.00000 rec=99.63%] ext[mse=0.00000 rec=99.53%] | cc=-1.40 rig=0.0000 a=0.027 env=True
epoch 33: int[mse=0.00000 rec=99.66%] ext[mse=0.00000 rec=99.58%] | cc=-1.40 rig=0.0000 a=0.027 env=True
epoch 34: int[mse=0.00000 rec=99.69%] ext[mse=0.00000 rec=99.61%] | cc=-1.40 rig=0.0000 a=0.027 env=True
epoch 35: int[mse=0.00000 rec=99.71%] ext[mse=0.00000 rec=99.63%] | cc=-1.40 rig=0.0000 a=0.027 env=True
epoch 36: int[mse=0.00000 rec=99.73%] ext[mse=0.00000 rec=99.65%] | cc=-1.39 rig=0.0000 a=0.027 env=True
epoch 37: int[mse=0.00000 rec=99.73%] ext[mse=0.00000 rec=99.67%] | cc=-1.39 rig=0.0000 a=0.027 env=True
epoch 38: int[mse=0.00000 rec=99.74%] ext[mse=0.00000 rec=99.67%] | cc=-1.39 rig=0.0000 a=0.027 env=True
epoch 39: int[mse=0.00000 rec=99.74%] ext[mse=0.00000 rec=99.67%] | cc=-1.39 rig=0.0000 a=0.027 env=True
best recovery — internal 99.74% | external 99.67% | ckpt sign_test_signed/sign_signed.pt
================================================================
SIGN TEST — best recovery (internal / external)
================================================================
lens_sign=canon internal=99.72% external=99.43%
lens_sign=signed internal=99.74% external=99.67%
The new tests show the external shell when aligned with the internal MSE using decoupled behavior, align cleanly to the new process of adjudication. This means the canon was flawed, and the sign-preserving structural boundaries are in fact the most utilizable state.
The alephs REQUIRE the signs to be preserved to retain the geometric structure, no exceptions.
==========================================================================================
[sign-test] lens_sign='canon'
==========================================================================================
geolip-svae-v2 TWO-RECON | battery(PatchSVAE) 52,131 + shell 44,204 + growth 0 | D_base4->D_lens16 V32 ps2 | cuda
battery -> INTERNAL recon (pure MSE) | shell -> EXTERNAL recon (detached stem, lens_sign=canon) | adam lr=0.001 wd=0 sched=onecycle
growth : parked (no stencil)
[ByteTrigramDataset] Loading corpus wikitext-2-raw-v1...
[ByteTrigramDataset] Corpus: 10,938,611 bytes (10.9 MB), 768 bytes/image, 14,242 non-overlapping images available (10,937,843 valid window starts)
epoch 0: int[mse=0.01549 rec=12.89%] ext[mse=0.02859 rec= 4.58%] | cc=-1.49 rig=0.0000 a=0.024 env=True
epoch 1: int[mse=0.00052 rec=30.14%] ext[mse=0.00300 rec=12.73%] | cc=-1.48 rig=0.0000 a=0.025 env=True
epoch 2: int[mse=0.00013 rec=47.44%] ext[mse=0.00070 rec=22.37%] | cc=-1.47 rig=0.0000 a=0.026 env=True
epoch 3: int[mse=0.00007 rec=62.95%] ext[mse=0.00032 rec=29.98%] | cc=-1.47 rig=0.0000 a=0.026 env=True
epoch 4: int[mse=0.00008 rec=16.89%] ext[mse=0.00024 rec=16.71%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 5: int[mse=0.00009 rec=57.83%] ext[mse=0.00025 rec=32.53%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 6: int[mse=0.00010 rec=62.00%] ext[mse=0.00029 rec=33.70%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 7: int[mse=0.00009 rec=50.16%] ext[mse=0.00032 rec=29.28%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 8: int[mse=0.00008 rec=33.23%] ext[mse=0.00031 rec=21.43%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 9: int[mse=0.00008 rec=29.09%] ext[mse=0.00031 rec=21.93%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 10: int[mse=0.00007 rec=43.08%] ext[mse=0.00030 rec=27.95%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 11: int[mse=0.00006 rec=48.34%] ext[mse=0.00026 rec=29.49%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 12: int[mse=0.00005 rec=44.14%] ext[mse=0.00023 rec=28.97%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 13: int[mse=0.00004 rec=36.71%] ext[mse=0.00021 rec=23.32%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 14: int[mse=0.00004 rec=51.07%] ext[mse=0.00019 rec=29.76%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 15: int[mse=0.00003 rec=55.61%] ext[mse=0.00017 rec=30.72%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 16: int[mse=0.00003 rec=82.98%] ext[mse=0.00016 rec=41.54%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 17: int[mse=0.00002 rec=63.17%] ext[mse=0.00014 rec=33.50%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 18: int[mse=0.00002 rec=66.96%] ext[mse=0.00013 rec=36.20%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 19: int[mse=0.00002 rec=75.74%] ext[mse=0.00012 rec=40.23%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 20: int[mse=0.00001 rec=81.11%] ext[mse=0.00010 rec=45.54%] | cc=-1.46 rig=0.0000 a=0.028 env=True
epoch 21: int[mse=0.00001 rec=69.71%] ext[mse=0.00009 rec=38.91%] | cc=-1.45 rig=0.0000 a=0.028 env=True
epoch 22: int[mse=0.00001 rec=89.51%] ext[mse=0.00008 rec=51.03%] | cc=-1.45 rig=0.0000 a=0.028 env=True
epoch 23: int[mse=0.00001 rec=93.21%] ext[mse=0.00006 rec=54.33%] | cc=-1.45 rig=0.0000 a=0.028 env=True
epoch 24: int[mse=0.00001 rec=74.30%] ext[mse=0.00006 rec=46.39%] | cc=-1.44 rig=0.0000 a=0.028 env=True
epoch 25: int[mse=0.00000 rec=96.38%] ext[mse=0.00005 rec=60.66%] | cc=-1.44 rig=0.0000 a=0.028 env=True
epoch 26: int[mse=0.00000 rec=98.80%] ext[mse=0.00004 rec=68.66%] | cc=-1.43 rig=0.0000 a=0.028 env=True
epoch 27: int[mse=0.00000 rec=99.20%] ext[mse=0.00003 rec=71.80%] | cc=-1.43 rig=0.0000 a=0.028 env=True
epoch 28: int[mse=0.00000 rec=99.37%] ext[mse=0.00003 rec=74.51%] | cc=-1.42 rig=0.0000 a=0.028 env=True
epoch 29: int[mse=0.00000 rec=99.32%] ext[mse=0.00002 rec=76.43%] | cc=-1.42 rig=0.0000 a=0.028 env=True
epoch 30: int[mse=0.00000 rec=99.53%] ext[mse=0.00002 rec=80.68%] | cc=-1.41 rig=0.0000 a=0.028 env=True
epoch 31: int[mse=0.00000 rec=99.56%] ext[mse=0.00002 rec=82.36%] | cc=-1.41 rig=0.0000 a=0.028 env=True
epoch 32: int[mse=0.00000 rec=99.62%] ext[mse=0.00002 rec=84.97%] | cc=-1.41 rig=0.0000 a=0.028 env=True
epoch 33: int[mse=0.00000 rec=99.65%] ext[mse=0.00001 rec=86.74%] | cc=-1.40 rig=0.0000 a=0.028 env=True
epoch 34: int[mse=0.00000 rec=99.68%] ext[mse=0.00001 rec=88.07%] | cc=-1.40 rig=0.0000 a=0.028 env=True
epoch 35: int[mse=0.00000 rec=99.70%] ext[mse=0.00001 rec=89.02%] | cc=-1.40 rig=0.0000 a=0.028 env=True
epoch 36: int[mse=0.00000 rec=99.71%] ext[mse=0.00001 rec=89.75%] | cc=-1.39 rig=0.0000 a=0.028 env=True
epoch 37: int[mse=0.00000 rec=99.72%] ext[mse=0.00001 rec=90.26%] | cc=-1.39 rig=0.0000 a=0.028 env=True
epoch 38: int[mse=0.00000 rec=99.73%] ext[mse=0.00001 rec=90.48%] | cc=-1.39 rig=0.0000 a=0.028 env=True
epoch 39: int[mse=0.00000 rec=99.73%] ext[mse=0.00001 rec=90.53%] | cc=-1.39 rig=0.0000 a=0.028 env=True
best recovery — internal 99.73% | external 90.53% | ckpt sign_test_canon/sign_canon.pt
==========================================================================================
[sign-test] lens_sign='signed'
==========================================================================================
geolip-svae-v2 TWO-RECON | battery(PatchSVAE) 52,131 + shell 44,204 + growth 0 | D_base4->D_lens16 V32 ps2 | cuda
battery -> INTERNAL recon (pure MSE) | shell -> EXTERNAL recon (detached stem, lens_sign=signed) | adam lr=0.001 wd=0 sched=onecycle
growth : parked (no stencil)
[ByteTrigramDataset] Loading corpus wikitext-2-raw-v1...
[ByteTrigramDataset] Corpus: 10,938,611 bytes (10.9 MB), 768 bytes/image, 14,242 non-overlapping images available (10,937,843 valid window starts)
epoch 0: int[mse=0.01550 rec=12.88%] ext[mse=0.01768 rec=10.14%] | cc=-1.49 rig=0.0000 a=0.024 env=True
epoch 1: int[mse=0.00053 rec=30.09%] ext[mse=0.00073 rec=21.73%] | cc=-1.48 rig=0.0000 a=0.025 env=True
epoch 2: int[mse=0.00013 rec=44.02%] ext[mse=0.00018 rec=38.30%] | cc=-1.47 rig=0.0000 a=0.025 env=True
epoch 3: int[mse=0.00008 rec=28.60%] ext[mse=0.00009 rec=30.84%] | cc=-1.47 rig=0.0000 a=0.025 env=True
epoch 4: int[mse=0.00008 rec=50.84%] ext[mse=0.00008 rec=49.70%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 5: int[mse=0.00009 rec=60.27%] ext[mse=0.00009 rec=47.82%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 6: int[mse=0.00009 rec=34.78%] ext[mse=0.00010 rec=33.67%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 7: int[mse=0.00009 rec=55.85%] ext[mse=0.00011 rec=43.34%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 8: int[mse=0.00009 rec=70.88%] ext[mse=0.00011 rec=57.15%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 9: int[mse=0.00008 rec=49.43%] ext[mse=0.00010 rec=43.24%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 10: int[mse=0.00007 rec=37.95%] ext[mse=0.00010 rec=36.92%] | cc=-1.46 rig=0.0000 a=0.025 env=True
epoch 11: int[mse=0.00006 rec=25.67%] ext[mse=0.00009 rec=27.67%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 12: int[mse=0.00005 rec=31.98%] ext[mse=0.00007 rec=28.85%] | cc=-1.46 rig=0.0000 a=0.026 env=True
epoch 13: int[mse=0.00004 rec=43.96%] ext[mse=0.00006 rec=41.25%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 14: int[mse=0.00004 rec=34.30%] ext[mse=0.00005 rec=31.27%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 15: int[mse=0.00003 rec=52.89%] ext[mse=0.00005 rec=44.82%] | cc=-1.46 rig=0.0000 a=0.027 env=True
epoch 16: int[mse=0.00003 rec=79.94%] ext[mse=0.00004 rec=68.72%] | cc=-1.46 rig=0.0000 a=0.028 env=True
epoch 17: int[mse=0.00002 rec=75.35%] ext[mse=0.00004 rec=61.30%] | cc=-1.46 rig=0.0000 a=0.028 env=True
epoch 18: int[mse=0.00002 rec=87.26%] ext[mse=0.00003 rec=75.07%] | cc=-1.46 rig=0.0000 a=0.028 env=True
epoch 19: int[mse=0.00002 rec=92.37%] ext[mse=0.00003 rec=80.44%] | cc=-1.46 rig=0.0000 a=0.029 env=True
epoch 20: int[mse=0.00001 rec=63.92%] ext[mse=0.00002 rec=52.35%] | cc=-1.46 rig=0.0000 a=0.029 env=True
epoch 21: int[mse=0.00001 rec=96.15%] ext[mse=0.00002 rec=86.80%] | cc=-1.45 rig=0.0000 a=0.029 env=True
epoch 22: int[mse=0.00001 rec=87.43%] ext[mse=0.00002 rec=76.03%] | cc=-1.45 rig=0.0000 a=0.029 env=True
epoch 23: int[mse=0.00001 rec=91.37%] ext[mse=0.00001 rec=78.56%] | cc=-1.45 rig=0.0000 a=0.029 env=True
epoch 24: int[mse=0.00001 rec=97.84%] ext[mse=0.00001 rec=91.84%] | cc=-1.44 rig=0.0000 a=0.030 env=True
epoch 25: int[mse=0.00000 rec=81.34%] ext[mse=0.00001 rec=68.67%] | cc=-1.44 rig=0.0000 a=0.030 env=True
epoch 26: int[mse=0.00000 rec=99.13%] ext[mse=0.00001 rec=96.60%] | cc=-1.43 rig=0.0000 a=0.030 env=True
epoch 27: int[mse=0.00000 rec=99.27%] ext[mse=0.00001 rec=97.22%] | cc=-1.43 rig=0.0000 a=0.030 env=True
epoch 28: int[mse=0.00000 rec=99.35%] ext[mse=0.00000 rec=97.58%] | cc=-1.42 rig=0.0000 a=0.030 env=True
epoch 29: int[mse=0.00000 rec=99.47%] ext[mse=0.00000 rec=98.30%] | cc=-1.42 rig=0.0000 a=0.030 env=True
epoch 30: int[mse=0.00000 rec=99.40%] ext[mse=0.00000 rec=97.67%] | cc=-1.41 rig=0.0000 a=0.030 env=True
epoch 31: int[mse=0.00000 rec=99.57%] ext[mse=0.00000 rec=98.77%] | cc=-1.41 rig=0.0000 a=0.030 env=True
epoch 32: int[mse=0.00000 rec=99.62%] ext[mse=0.00000 rec=98.93%] | cc=-1.41 rig=0.0000 a=0.030 env=True
epoch 33: int[mse=0.00000 rec=99.65%] ext[mse=0.00000 rec=99.07%] | cc=-1.40 rig=0.0000 a=0.030 env=True
epoch 34: int[mse=0.00000 rec=99.68%] ext[mse=0.00000 rec=99.17%] | cc=-1.40 rig=0.0000 a=0.030 env=True
epoch 35: int[mse=0.00000 rec=99.70%] ext[mse=0.00000 rec=99.22%] | cc=-1.40 rig=0.0000 a=0.030 env=True
epoch 36: int[mse=0.00000 rec=99.72%] ext[mse=0.00000 rec=99.27%] | cc=-1.39 rig=0.0000 a=0.030 env=True
epoch 37: int[mse=0.00000 rec=99.73%] ext[mse=0.00000 rec=99.29%] | cc=-1.39 rig=0.0000 a=0.030 env=True
epoch 38: int[mse=0.00000 rec=99.73%] ext[mse=0.00000 rec=99.31%] | cc=-1.39 rig=0.0000 a=0.030 env=True
epoch 39: int[mse=0.00000 rec=99.74%] ext[mse=0.00000 rec=99.31%] | cc=-1.39 rig=0.0000 a=0.030 env=True
best recovery — internal 99.74% | external 99.31% | ckpt sign_test_signed/sign_signed.pt
================================================================
SIGN TEST — best recovery (internal / external)
================================================================
lens_sign=canon internal=99.73% external=90.53%
lens_sign=signed internal=99.74% external=99.31%
external delta (signed - canon): +8.78% -> SIGNS CARRY CONTENT — hypothesis supported
I believe the Transformer is 2 steps back and 1 step forward, rather than the 3 steps forward I was anticipating. I'll need to enter blueprint stages again with the new eliminations in mind while accounting for the various successes.
One approach I believe is valuable, is the projection capacity and the rigid encapsulant folding. This is most definitely a direction that needs to be explored.
Even so there were successes in the mix.
Article wasn't ready yet, the experiments for this round need more work on each before I can attest to certain rules, and the wording needs considerable amounts of work.
The article will be out when the experiments are ready.
Bert trainer converged as well now when bert systems are fed as trigrams as well.
geolip-svae-bert | lens=single ladder [4, 8, 16, 32] D_dec=32 | void=on spectral=on(2L) | V32 ps2 | params 143,671 | cuda
trainer: adamw lr=0.001 wd=0.0001 clip=1.0 | rigid_hinge=3.0 (margin 0.25·crit) diff=0.2 (ramp 30%) recon=cosine_sim | sched=onecycle | workers=4
[BERTVectorDataset] Loading corpus wikitext-2-raw-v1...
[BERTVectorDataset] 23,767 non-empty lines; encoding via BERT...
[BERT] loading bert-base-uncased on cuda...
Loading weights: 100%
 199/199 [00:00<00:00, 4028.39it/s, Materializing param=pooler.dense.weight]
BertModel LOAD REPORT from: bert-base-uncased
Key | Status | |
-------------------------------------------+------------+--+-
cls.seq_relationship.bias | UNEXPECTED | |
cls.predictions.bias | UNEXPECTED | |
cls.seq_relationship.weight | UNEXPECTED | |
cls.predictions.transform.dense.bias | UNEXPECTED | |
cls.predictions.transform.LayerNorm.bias | UNEXPECTED | |
cls.predictions.transform.LayerNorm.weight | UNEXPECTED | |
cls.predictions.transform.dense.weight | UNEXPECTED | |
Notes:
- UNEXPECTED :can be ignored when loading from different task/architecture; not ok if you expect identical arch.
[BERT] 100,000 unit-normalized vectors in 1.7s
[BERTVectorDataset] 100,000 BERT vectors held in CPU memory (307.2MB)
[eval_corpus] loading Trelis/tiny-shakespeare (held-out, unrelated to training)...
[eval_corpus] Trelis/tiny-shakespeare: 472 lines, encoding via BERT...
[BERT] loading bert-base-uncased on cuda:0...
Loading weights: 100%
 199/199 [00:00<00:00, 4059.70it/s, Materializing param=pooler.dense.weight]
BertModel LOAD REPORT from: bert-base-uncased
Key | Status | |
-------------------------------------------+------------+--+-
cls.seq_relationship.bias | UNEXPECTED | |
cls.predictions.bias | UNEXPECTED | |
cls.seq_relationship.weight | UNEXPECTED | |
cls.predictions.transform.dense.bias | UNEXPECTED | |
cls.predictions.transform.LayerNorm.bias | UNEXPECTED | |
cls.predictions.transform.LayerNorm.weight | UNEXPECTED | |
cls.predictions.transform.dense.weight | UNEXPECTED | |
Notes:
- UNEXPECTED :can be ignored when loading from different task/architecture; not ok if you expect identical arch.
[BERT] 8,192 unit-normalized vectors in 0.1s
epoch 0: mse=0.00059 | rand=99.05% nat=99.06% | alpha=0.0239 | dev=0.0047 in_env=True
epoch 1: mse=0.00001 | rand=99.88% nat=99.88% | alpha=0.0239 | dev=0.0037 in_env=True
epoch 2: mse=0.00001 | rand=99.90% nat=99.87% | alpha=0.0239 | dev=0.0041 in_env=True
epoch 3: mse=0.00001 | rand=98.74% nat=98.71% | alpha=0.0240 | dev=0.0032 in_env=True
epoch 4: mse=0.00001 | rand=99.54% nat=99.53% | alpha=0.0244 | dev=0.0031 in_env=True
epoch 5: mse=0.00001 | rand=99.53% nat=99.52% | alpha=0.0248 | dev=0.0032 in_env=True
epoch 6: mse=0.00001 | rand=99.81% nat=99.81% | alpha=0.0251 | dev=0.0025 in_env=True
epoch 7: mse=0.00000 | rand=99.93% nat=99.93% | alpha=0.0254 | dev=0.0028 in_env=True
epoch 8: mse=0.00000 | rand=99.82% nat=99.82% | alpha=0.0256 | dev=0.0030 in_env=True
epoch 9: mse=0.00000 | rand=99.97% nat=99.97% | alpha=0.0257 | dev=0.0034 in_env=True
epoch 10: mse=0.00000 | rand=99.97% nat=99.97% | alpha=0.0259 | dev=0.0033 in_env=True
epoch 11: mse=0.00000 | rand=99.98% nat=99.98% | alpha=0.0259 | dev=0.0033 in_env=True
epoch 12: mse=0.00000 | rand=99.99% nat=99.99% | alpha=0.0259 | dev=0.0031 in_env=True
epoch 13: mse=0.00000 | rand=99.99% nat=99.99% | alpha=0.0259 | dev=0.0028 in_env=True
epoch 14: mse=0.00000 | rand=100.00% nat=100.00% | alpha=0.0259 | dev=0.0029 in_env=True
epoch 15: mse=0.00000 | rand=100.00% nat=100.00% | alpha=0.0259 | dev=0.0031 in_env=True
epoch 16: mse=0.00000 | rand=100.00% nat=100.00% | alpha=0.0259 | dev=0.0033 in_env=True
epoch 17: mse=0.00000 | rand=100.00% nat=100.00% | alpha=0.0259 | dev=0.0034 in_env=True
epoch 18: mse=0.00000 | rand=100.00% nat=100.00% | alpha=0.0259 | dev=0.0034 in_env=True
epoch 19: mse=0.00000 | rand=100.00% nat=100.00% | alpha=0.0259 | dev=0.0034 in_env=True
best cos recovery: 100.00% | checkpoint: geolip_svae_bert_results/geolip_svae_bert.pt
https://huggingface.co/AbstractPhil/geolip-svae-transformer/blob/main/transformer_v3.py
Lens setting: "crusher"
The crusher is live, a simple manifestation with a large punch to the statistics capacity. They are essentially pulverizing the information into a more compact shape and learning using it, simple manifestation that converges like the standard SVAE. Requires a bit more work, and a couple of the math faults still exist in the model that I'm working through. The current crusher requires optimizations and tweaks to the formula to speed it up.
The problem with these particular faults is larger than simple solution, the theorems around them require too much math. The approximations require a compact solution.
The crusher isn't quite there yet. I'll post the results for a converging crusher with better optimization than the current asap.
For single the current formula snaps a 100% so that's good.
geolip-svae-transformer | lens=single ladder [4, 8, 16] D_dec=16 | void=on spectral=on(2L) | V32 ps2 | params 96,849 | cuda
trainer: adamw lr=0.001 wd=0.0001 clip=1.0 | rigid_hinge=3.0 (margin 0.25·crit) diff=0.2 (ramp 30%) recon=pure_MSE | sched=onecycle | workers=8
torch.compile ON (mode=reduce-overhead) — first step pays compile cost
[ByteTrigramDataset] Loading corpus wikitext-2-raw-v1...
[ByteTrigramDataset] Corpus: 10,938,611 bytes (10.9 MB), 768 bytes/image, 14,242 non-overlapping images available (10,937,843 valid window starts)
epoch 0: mse=0.03261 | rand= 4.36% nat= 3.00% | alpha=0.0244 | dev=0.0085 in_env=True
epoch 1: mse=0.00302 | rand=25.08% nat=23.79% | alpha=0.0253 | dev=0.0066 in_env=True
epoch 2: mse=0.00027 | rand=32.96% nat=35.06% | alpha=0.0253 | dev=0.0058 in_env=True
epoch 3: mse=0.00027 | rand=42.94% nat=44.58% | alpha=0.0252 | dev=0.0042 in_env=True
epoch 4: mse=0.00027 | rand=19.85% nat=21.32% | alpha=0.0250 | dev=0.0045 in_env=True
epoch 5: mse=0.00020 | rand= 9.62% nat=11.55% | alpha=0.0249 | dev=0.0048 in_env=True
epoch 6: mse=0.00014 | rand=34.38% nat=39.38% | alpha=0.0250 | dev=0.0040 in_env=True
epoch 7: mse=0.00011 | rand=20.90% nat=22.83% | alpha=0.0251 | dev=0.0038 in_env=True
epoch 8: mse=0.00009 | rand=49.56% nat=53.20% | alpha=0.0253 | dev=0.0039 in_env=True
epoch 9: mse=0.00008 | rand=24.78% nat=28.64% | alpha=0.0256 | dev=0.0039 in_env=True
epoch 10: mse=0.00007 | rand=43.96% nat=46.96% | alpha=0.0258 | dev=0.0041 in_env=True
epoch 11: mse=0.00006 | rand=53.64% nat=54.76% | alpha=0.0260 | dev=0.0041 in_env=True
epoch 12: mse=0.00005 | rand=47.99% nat=48.25% | alpha=0.0263 | dev=0.0039 in_env=True
epoch 13: mse=0.00004 | rand=36.85% nat=40.04% | alpha=0.0265 | dev=0.0036 in_env=True
epoch 14: mse=0.00004 | rand=65.54% nat=68.46% | alpha=0.0267 | dev=0.0035 in_env=True
epoch 15: mse=0.00003 | rand=89.39% nat=92.22% | alpha=0.0269 | dev=0.0034 in_env=True
epoch 16: mse=0.00003 | rand=42.52% nat=44.34% | alpha=0.0270 | dev=0.0033 in_env=True
epoch 17: mse=0.00002 | rand=79.55% nat=83.27% | alpha=0.0272 | dev=0.0033 in_env=True
epoch 18: mse=0.00002 | rand=66.53% nat=69.91% | alpha=0.0273 | dev=0.0032 in_env=True
epoch 19: mse=0.00002 | rand=42.65% nat=42.95% | alpha=0.0274 | dev=0.0031 in_env=True
epoch 20: mse=0.00001 | rand=94.95% nat=96.85% | alpha=0.0275 | dev=0.0030 in_env=True
epoch 21: mse=0.00001 | rand=62.79% nat=65.49% | alpha=0.0276 | dev=0.0031 in_env=True
epoch 22: mse=0.00001 | rand=97.06% nat=98.77% | alpha=0.0276 | dev=0.0031 in_env=True
epoch 23: mse=0.00001 | rand=93.82% nat=95.22% | alpha=0.0276 | dev=0.0030 in_env=True
epoch 24: mse=0.00001 | rand=97.94% nat=99.36% | alpha=0.0277 | dev=0.0029 in_env=True
epoch 25: mse=0.00000 | rand=95.35% nat=96.22% | alpha=0.0277 | dev=0.0029 in_env=True
epoch 26: mse=0.00000 | rand=98.93% nat=99.65% | alpha=0.0277 | dev=0.0029 in_env=True
epoch 27: mse=0.00000 | rand=98.65% nat=99.59% | alpha=0.0277 | dev=0.0029 in_env=True
epoch 28: mse=0.00000 | rand=99.18% nat=99.65% | alpha=0.0277 | dev=0.0028 in_env=True
epoch 29: mse=0.00000 | rand=99.27% nat=99.85% | alpha=0.0277 | dev=0.0028 in_env=True
epoch 30: mse=0.00000 | rand=99.31% nat=99.69% | alpha=0.0277 | dev=0.0028 in_env=True
epoch 31: mse=0.00000 | rand=99.43% nat=100.00% | alpha=0.0277 | dev=0.0028 in_env=True
epoch 32: mse=0.00000 | rand=99.47% nat=100.00% | alpha=0.0277 | dev=0.0028 in_env=True
epoch 33: mse=0.00000 | rand=99.51% nat=100.00% | alpha=0.0277 | dev=0.0028 in_env=True
epoch 34: mse=0.00000 | rand=99.54% nat=100.00% | alpha=0.0277 | dev=0.0028 in_env=True
epoch 35: mse=0.00000 | rand=99.56% nat=100.00% | alpha=0.0277 | dev=0.0027 in_env=True
epoch 36: mse=0.00000 | rand=99.58% nat=100.00% | alpha=0.0277 | dev=0.0027 in_env=True
epoch 37: mse=0.00000 | rand=99.59% nat=100.00% | alpha=0.0277 | dev=0.0027 in_env=True
epoch 38: mse=0.00000 | rand=99.60% nat=100.00% | alpha=0.0277 | dev=0.0027 in_env=True
epoch 39: mse=0.00000 | rand=99.60% nat=100.00% | alpha=0.0277 | dev=0.0027 in_env=True
best byte recovery: 99.60% | checkpoint: geolip_svae_transformer_results/geolip_svae_transformer.pt
adjudication (text -> model -> text):
'the cat sat on the mat' -> 'the cat sat on the mat'
'machine learning' -> 'machine learning'
'hello world' -> 'hello world'
Will need to attach unrelated data for recon testing to increase the nat difficulty. It's too easy.
Alright I hooked up Trelis/tiny-shakespeare for independent eval.
Fresh eyes of the day I see a couple of fairly crucial mistakes, I'll get them worked out.
Should work, it's a little unstable still but it gets there.
The explicit rigidity is converging into implicit behavior, and with that the model can recon random noise and the actual behavior of the byte math.
Multiscale lensed bytewise transformation through a rigid geometric lens.
geolip-svae-transformer | lens=single ladder [4, 8, 16] D_dec=16 | void=on spectral=on(2L) | V32 ps2 | params 96,849 | cuda
trainer: AdamW lr=0.0001 clip=1.0 | rigid_hinge=3.0 (margin 0.25·crit) diff=0.2 (ramp 30%) recon=pure_MSE | OneCycle | workers=4
[ByteTrigramDataset] Loading corpus wikitext-2-raw-v1...
[ByteTrigramDataset] Corpus: 10,938,611 bytes (10.9 MB), 768 bytes/image, 14,242 non-overlapping images available (10,937,843 valid window starts)
epoch 0: mse=0.10583 | rand= 1.39% nat= 0.79% | alpha=0.0239 | dev=0.0143 in_env=True
epoch 1: mse=0.03172 | rand= 2.63% nat= 1.21% | alpha=0.0240 | dev=0.0139 in_env=True
epoch 2: mse=0.01195 | rand= 4.19% nat= 4.27% | alpha=0.0244 | dev=0.0132 in_env=True
epoch 3: mse=0.00322 | rand=16.88% nat=15.48% | alpha=0.0249 | dev=0.0071 in_env=True
epoch 4: mse=0.00045 | rand=27.95% nat=29.86% | alpha=0.0251 | dev=0.0059 in_env=True
epoch 5: mse=0.00021 | rand=37.33% nat=37.96% | alpha=0.0252 | dev=0.0058 in_env=True
epoch 6: mse=0.00012 | rand=38.58% nat=38.66% | alpha=0.0253 | dev=0.0055 in_env=True
epoch 7: mse=0.00009 | rand=41.14% nat=40.70% | alpha=0.0253 | dev=0.0054 in_env=True
epoch 8: mse=0.00006 | rand=54.71% nat=55.22% | alpha=0.0254 | dev=0.0052 in_env=True
epoch 9: mse=0.00005 | rand=66.73% nat=67.98% | alpha=0.0254 | dev=0.0050 in_env=True
epoch 10: mse=0.00004 | rand=74.02% nat=74.04% | alpha=0.0254 | dev=0.0050 in_env=True
epoch 11: mse=0.00003 | rand=72.36% nat=72.56% | alpha=0.0254 | dev=0.0049 in_env=True
epoch 12: mse=0.00003 | rand=68.90% nat=67.53% | alpha=0.0254 | dev=0.0048 in_env=True
epoch 13: mse=0.00003 | rand=76.69% nat=77.38% | alpha=0.0254 | dev=0.0046 in_env=True
epoch 14: mse=0.00002 | rand=73.02% nat=72.70% | alpha=0.0254 | dev=0.0045 in_env=True
epoch 15: mse=0.00002 | rand=70.31% nat=69.88% | alpha=0.0254 | dev=0.0044 in_env=True
epoch 16: mse=0.00002 | rand=79.93% nat=77.46% | alpha=0.0254 | dev=0.0043 in_env=True
epoch 17: mse=0.00002 | rand=76.01% nat=73.98% | alpha=0.0255 | dev=0.0048 in_env=True
epoch 18: mse=0.00002 | rand=83.62% nat=82.27% | alpha=0.0255 | dev=0.0055 in_env=True
epoch 19: mse=0.00001 | rand=87.61% nat=88.60% | alpha=0.0255 | dev=0.0051 in_env=True
epoch 20: mse=0.00001 | rand=87.44% nat=88.66% | alpha=0.0255 | dev=0.0049 in_env=True
epoch 21: mse=0.00001 | rand=89.98% nat=87.82% | alpha=0.0255 | dev=0.0047 in_env=True
epoch 22: mse=0.00001 | rand=91.60% nat=89.77% | alpha=0.0255 | dev=0.0046 in_env=True
epoch 23: mse=0.00001 | rand=92.70% nat=91.50% | alpha=0.0255 | dev=0.0045 in_env=True
epoch 24: mse=0.00001 | rand=92.76% nat=91.87% | alpha=0.0255 | dev=0.0045 in_env=True
epoch 25: mse=0.00001 | rand=93.11% nat=91.68% | alpha=0.0255 | dev=0.0044 in_env=True
epoch 26: mse=0.00001 | rand=94.42% nat=94.23% | alpha=0.0255 | dev=0.0044 in_env=True
epoch 27: mse=0.00001 | rand=94.46% nat=93.55% | alpha=0.0255 | dev=0.0043 in_env=True
epoch 28: mse=0.00001 | rand=95.02% nat=94.58% | alpha=0.0255 | dev=0.0043 in_env=True
epoch 29: mse=0.00001 | rand=93.72% nat=94.69% | alpha=0.0255 | dev=0.0043 in_env=True
epoch 30: mse=0.00001 | rand=90.87% nat=88.89% | alpha=0.0256 | dev=0.0035 in_env=True
epoch 31: mse=0.00001 | rand=90.68% nat=89.57% | alpha=0.0256 | dev=0.0032 in_env=True
epoch 32: mse=0.00001 | rand=94.78% nat=94.39% | alpha=0.0256 | dev=0.0030 in_env=True
epoch 33: mse=0.00001 | rand=95.82% nat=95.86% | alpha=0.0256 | dev=0.0030 in_env=True
epoch 34: mse=0.00000 | rand=96.34% nat=96.15% | alpha=0.0256 | dev=0.0031 in_env=True
epoch 35: mse=0.00000 | rand=96.59% nat=96.65% | alpha=0.0256 | dev=0.0031 in_env=True
epoch 36: mse=0.00000 | rand=96.74% nat=96.80% | alpha=0.0256 | dev=0.0031 in_env=True
epoch 37: mse=0.00000 | rand=96.83% nat=96.86% | alpha=0.0256 | dev=0.0031 in_env=True
epoch 38: mse=0.00000 | rand=96.87% nat=96.86% | alpha=0.0256 | dev=0.0031 in_env=True
epoch 39: mse=0.00000 | rand=96.87% nat=96.65% | alpha=0.0256 | dev=0.0031 in_env=True
best byte recovery: 96.87% | checkpoint: geolip_svae_transformer_results/geolip_svae_transformer.pt
adjudication (text -> model -> text):
'the cat sat on the mat' -> 'the cat sat on tie mat'
'machine learning' -> 'machine learning'
'hello world' -> 'hello world'
Perfect? Not yet. Faster than SVAE battery convergence? Most definitely.
The optimization on the centrifuge was quite lackluster. The hardware doesn't support such behavior, so I've continued to the crusher.
The transformer is operational, which takes the behavior of the H2 battery and directly projects it to a multiscale structure. Roughly turns 57k params to around 90k params, and with this behavior the model converges SEMI-CLOSE to the SVAE current spectrum in considerably less epochs. So stay tuned on that one, the transformer did converge.
I've implanted a rigid formula that allows this direct behavior from the H2 battery to superimpose onto adjacent structural boundaries, and with that built aleph and void into the system as well. These are guarantees.
Four lenses was too much when running a quaternion bank to handle such complex interactions reasonably, so I will need to work something out in the future to get a full centrifuge system working.
The crusher however is essentially a guarantee. There's a definite valuation set that can be easily obtained by simply pulverizing and analyzing the remains.
I have the aleph formula.
After tomorrow I'm taking a few days to recover from overwork and relax, so over the weekend I'll likely set up a few experiment sweeps or run some sweeps on the weekend if I don't start them on Monday or Tuesday.
Burning the daytime oil and midnight oil has been taxing, I need to relax for a bit.
I believe I've discovered the first needs-based component from this structure.
Centrifuged dissonance rupture sampling.
Magnitude testing with ripter shows multiple reduced magnitudes will coalesce less infinites and still converge, while the squared magnitude standard will introduce a type of unilaterally conjunctive series of ripter identified infinites.
For the baseline H2 trigram training:
The adjudication principality of compartmentalized information shows those infinites are nearly fully encompassing the state of the model, they are 22 identifiable omegas with only one finite and two persistent features, while the structure when analyzed shows the structure is predominantly nearly fully utilized. The only way for this to be possible is if the model is directly utilizing the infinites to make decisions.
I have a working hypothesis based on omegas created by the H2 battery.
So far I've identified four types of potential Omega states using ripter which have been codified.
Using the default config magnitude of 2 on the baseline trigram battery we have generated;
"persistent_infinity_field" as the codified response.
While pi magnitude sings a different tune entirely.
"rupture_coalescence_field",
"persistent_infinity_field": (
"Most H0 classes remain infinite under the threshold; "
"axis residents are mostly separated and persistent."
),
"rupture_coalescence_field": (
"Many axes persist, but several finite H0 deaths indicate "
"controlled local coalescence/rupture."
),
"infinity_pair_field": (
"Mostly persistent infinity axes with small pair components."
),
"finite_carrier_field": (
"Finite merge activity is high enough to indicate a carrier "
"rather than pure infinity field."
),
Each is indicated to be a potential omega and this will require heavy ablation to determine if this is true or grasping at noise.
Transformers default sentencepiece for the t5-base can't be manually swapped to the fast version, I had to manually load a fast version of sentencepiece rust to actually make use of it.
This was an inconvenience that cost quite a bit of debugging time, but now that it's resolved I can start the full trains.
Prelim 50 epoch shows full recon is possible.
The most confused pairs may have a mathematical rounding fault to be addressed, I'll figure that out when the 100 epoch finishes.
Organizing the branch system:
I didn't like my first design, I'll be keeping the prototypes in a separate package in the same repo. This is cleaner.
geolip_svae
svae_proto
Basically separate readmes, separate requirements, and so on. This will be cleaner and will scale naturally.
Musings:
In a vain attempt to keep things organized, I'll be segmenting experiments to another branch, but I need to be strict about my own behavior. Forming an experimental branch won't matter at this point if I can't keep the repos organized, much of the repo WILL BE experimental prototypes within the geolip_svae.prototypes package - but without any explicit guards for my own behavior. I need to clean things up a bit and prevent my own regular habits from causing chaos.
I've done this before, where I go off on huge experiment lines and create massive tangents of 20-200 experiments, so I'll be forming an experimental branch which uses only the "main" geolip_svae package core model packages and code, while existing in the same experimental repo.
So the core code for the model, the behaviors, the optimization, and the expectations will exist in the primary "main" branch, and the "experimental" branch will house the incomplete, untested, or unrefined prototypes.
Merging into the main will be a modularization process, likely done by hand or using Claude to assist, and then the main code updated in the experimental while preserving the prototypes and their separation. This is a stop gap for me, because I notoriously climb into rabbit holes of experiment - only to find something interesting, but in the stupor I find the common cork board room full of yarn. It never starts out a cork board room, it just evolves into one naturally.
Trigram 128 converged cleanly by epoch 10 with 4096 patches. 100% byte recon, >99.9% recon for trigrams.
I set up a vocabulary/token experiment to probe the capacity for full tokenization, that's running next.
One thing to note. The trigram variants are behaving like miniature decoder llms more than they are encoders.
The energy concentration is internal decoder-centric, so this may potentially be a symptom of an emergent internal energy weighting spectrum that may require additional alpha. More tests will be required. From a glance, I can only guess. The curve is clear, decreasing S0 increasing SD throughout the train.
The symptom is present that this model is decoder-weighted, which could mean many things. They are self-encapsulated adjudication units so they will work, but this gives me a series of ideas based on the converged pattern to better orient the internals using a few continuum topology methods experimented on in the past.
These were experimental continuum rotators. They rotate a feature away from latent space, process there, and then rotate the result back. Early experimentation was yielding, but slow, which is why I moved in a different direction. The math can help explain some of the symptoms here I believe.
The naming schema for "omega" is not to be confused with this variation. That was an experimental attempt to curate an omega pathway that could yield, but more often produced NaN so it had limited scope in the gradients.
Upon review I'm thinking it might not contribute much. So I'll need to consult the continuum research closer.
Anyway this isn't important currently, tokenization is. I'll report later today with the findings.