Zum Inhalt springen

ModelAufgelistetExtern

Geolip Aleph Qwen 3.5 0.8b Instruct

This is the research record. If you're here for something usable:

Quellehuggingface.co/AbstractPhil/geolip-aleph-qwen-3.5-0.8b-instruct/tree/f622993e346f46e86bfdd5b631c1c1c56c40d752

ArtOtherBasismodellQwen/Qwen3.5-0.8BVersionRevision f622993DBewertung

Modellquelle

Art: Other
Dateigröße: 36,9 KB–126 MB
Basismodell: Qwen/Qwen3.5-0.8B
Version: vRevision f622993
Aktualisiert: 14. Sept. 2026
Downloads: 0
Quelle: Hugging Face

Quellenauszug

This is the research record. If you're here for something usable:

Quellen

1 Quelle

Hugging FacePrimärVersion f622993e346f46e86bfdd5b631c1c1c56c40d752

huggingface.co/AbstractPhil/geolip-aleph-qwen-3.5-0.8b-instruct/tree/f622993e346f46e86bfdd5b631c1c1c56c40d752

Verifiziert 9. Aug.

Modellartefakte

32 Artefakte

caption_deliverable/v35e4_caption_full_s0.pt

pt · 24,0 MB · SHA-256 9c559c9c56f1…cf63 · Hugging Face

caption_deliverable/v35e4_caption_full_s1.pt

pt · 24,0 MB · SHA-256 02dd5ff698a4…676b · Hugging Face

eval_slices_v35.pt

pt · 20,5 MB · SHA-256 37743c018eb9…7361 · Hugging Face

exp001_placement/ckpt/v35e1_all24_s0.pt

pt · 24,0 MB · SHA-256 67aa2af638ea…b116 · Hugging Face

exp001_placement/ckpt/v35e1_full6_s0.pt

pt · 6,00 MB · SHA-256 9c9cc342dfad…bd1a · Hugging Face

exp001_placement/ckpt/v35e1_linear18_s0.pt

pt · 18,0 MB · SHA-256 a600805d8429…2fb7 · Hugging Face

exp002_registers/ckpt/v35e2_all24_s0.pt

pt · 24,0 MB · SHA-256 0fbadab0ee4c…f93e · Hugging Face

exp002_registers/ckpt/v35e2_all24_s1.pt

pt · 24,0 MB · SHA-256 fd6494d310e8…e202 · Hugging Face

exp003_termination/ckpt/v35e3_packed_s0.pt

pt · 24,0 MB · SHA-256 698b4a3a0871…25fb · Hugging Face

exp003_termination/ckpt/v35e3_rows_s0.pt

pt · 24,0 MB · SHA-256 bde5273bdac5…0b62 · Hugging Face

exp004_caption/ckpt/v35e4_caption_full_s0.pt

pt · 24,0 MB · SHA-256 9c559c9c56f1…cf63 · Hugging Face

exp004_caption/ckpt/v35e4_caption_full_s1.pt

pt · 24,0 MB · SHA-256 02dd5ff698a4…676b · Hugging Face

exp005_specialists/ckpt/v35e5_bbox_grounding_s0.pt

pt · 24,0 MB · SHA-256 1260a1935cb5…bdd8 · Hugging Face

exp005_specialists/ckpt/v35e5_depth_analysis_s0.pt

pt · 24,0 MB · SHA-256 ea0ec5a214a9…febf · Hugging Face

exp005_specialists/ckpt/v35e5_subject_fixation_s0.pt

pt · 24,0 MB · SHA-256 d227b572ec6c…23b5 · Hugging Face

exp006_math/ckpt/v35e6_math_s0.pt

pt · 24,0 MB · SHA-256 e52685208227…43cd · Hugging Face

exp007_collective/ckpt/v35e7_collective_s0.pt

pt · 126 MB · SHA-256 a02f89fb1322…8811 · Hugging Face

exp007_collective/ckpt/v35e7_collective_s1.pt

pt · 126 MB · SHA-256 e55ff03e91be…93d0 · Hugging Face

exp009_monolith/ckpt/v35e9_monolith_s0.pt

pt · 24,0 MB · SHA-256 29f0e470bedd…95e8 · Hugging Face

exp010_controller/ckpt/v35e7_collective_s0.pt

pt · 126 MB · SHA-256 a02f89fb1322…8811 · Hugging Face

exp013_experts/ckpt/v35e13_algebra_direct_s0.pt

pt · 24,0 MB · SHA-256 74e0b273dba2…45d1 · Hugging Face

exp013_experts/ckpt/v35e13_algebra_steps_s0.pt

pt · 24,0 MB · SHA-256 0e4a478bf94b…8050 · Hugging Face

exp013_experts/ckpt/v35e13_algebra_steps_s1.pt

pt · 24,0 MB · SHA-256 2ae28ac40af8…c616 · Hugging Face

exp013_experts/ckpt/v35e13_arith_multi_direct_s0.pt

pt · 24,0 MB · SHA-256 53b65bb7c12d…c61b · Hugging Face

exp013_experts/ckpt/v35e13_arith_multi_steps_s0.pt

pt · 24,0 MB · SHA-256 d69a9681b597…c0cb · Hugging Face

exp013_experts/ckpt/v35e13_arith_multi_steps_s1.pt

pt · 24,0 MB · SHA-256 6d4d25c3a492…ffcf · Hugging Face

exp013_experts/ckpt/v35e13_baseconv_direct_s0.pt

pt · 24,0 MB · SHA-256 7944ebde4c57…77f7 · Hugging Face

exp013_experts/ckpt/v35e13_fractions_direct_s0.pt

pt · 24,0 MB · SHA-256 768d718ea230…7a10 · Hugging Face

exp013_experts/ckpt/v35e13_fractions_direct_s1.pt

pt · 24,0 MB · SHA-256 0f58b7cb041f…1de0 · Hugging Face

exp013_experts/ckpt/v35e13_sequences_direct_s0.pt

pt · 24,0 MB · SHA-256 6364f7ea3830…c5e6 · Hugging Face

exp014_basins/ckpt/v35e14_keys_s0.pt

pt · 36,9 KB · SHA-256 ad3670679b11…192f · Hugging Face

exp014_basins/ckpt/v35e14_keysb_directalg_s0.pt

pt · 37,3 KB · SHA-256 72121c5a4d6c…1886 · Hugging Face

Quellenauszüge

2 Auszüge

Readme9. Aug.huggingface.co/AbstractPhil/geolip-aleph-qwen-3.5-0.8b-instruct

--- license: apache-2.0 base_model: Qwen/Qwen3.5-0.8B tags: - aleph - adapters - mixture-of-experts - second-model-validation --- # geolip-aleph-qwen-3.5-0.8b-instruct **This is the research record.** If you're here for something usable: | you want | go to | |---|---| | a working caption adapter for Qwen3.5-0.8B (validated both seeds, standalone loader) | [qwen3.5-0.8b-relay-caption](https://huggingface.co/AbstractPhil/qwen3.5-0.8b-relay-caption) | | the framework — train / attach / align / detach these adapters yourself | [amoe-lora](https://huggingface.co/AbstractPhil/amoe-lora) | | the full experimental story with citations | `article_ft3.md` in this repo | | the map of every related repo | `REPO_MAP.md` in this repo | **The scientific headline (n=48 matrices; 2-seed on the replicated verdicts):** always-on single-task adapters on this trunk are *mutually destructive* — the depth specialist alone reduces captioning to token-F1 0.0014 with zero terminations — and the aleph dispatch **rescues specialize-regime anchors almost completely** (caption returns to within 0.02 of gauge under every specialist mask) while **failing blend-regime anchors entirely** (the math experts escape damping, ratios 0.86–1.6 vs the ≥3× target, and measurably trample chain-of-thought the trunk handles natively). **Regime — not machinery — decides whether dispatch helps**, and the probe layer reads the regime in advance (directionally) for the image family. That one law is measured at four scales across exp011/014/017/018. **Reading order:** the zero-shot fingerprint below → the experiments table (each row links a self-contained package with raw ledgers and self-asserting results) → the dated amendment blocks inside exp005–014 and exp021. The amendments, a formal retraction, and a corrected overclaim are not blemishes; they are the method — every package was adversarially audited (several retroactively), and the audits' corrections shipped as dated blocks inside the packages. --- Second-model validation of the [geolip-aleph-qwen](https://huggingface.co/AbstractPhil/geolip-aleph-qwen) program on **Qwen/Qwen3.5-0.8B** — a dual-tower VLM whose LLM tower is a **hybrid**: 18 Gated-DeltaNet linear-attention blocks + 6 full-attention blocks (indices 3,7,11,15,19,23), d=1024, 248,320-token vocab, tied embeddings. Every law certified on the Qwen2....

Note9. Aug.huggingface.co/AbstractPhil/geolip-aleph-qwen-3.5-0.8b-instruct

Source context: 0 downloads · 1 likes · Repo AbstractPhil/geolip-aleph-qwen-3.5-0.8b-instruct