augmem/ES-AIST-81M-preview-GGUF

ES-AIST-81M Preview GGUF

This repository contains GGUF quantizations of augmem/ES-AIST-81M-preview.

Base model:

• augmem/ES-AIST-81M-preview

Quantizations:

• ES-AIST-81M_q8_0.gguf
• ES-AIST-81M_q5_1.gguf

These files were produced from:

• source checkpoint: es_aist_full_v13_anchor_memory_eventboost_er125_bs4096_nw0_l4b/best_model.pt
• exported checkpoint epoch: 6

Exact Release Metrics

The quantized files correspond to the same release checkpoint and eval bundle as the base repo.

Scoped status:

• This checkpoint passes the local ES-AIST memory/entity-signal gate for compact open AIST models in this release line.
• The claim is limited to the memory-oriented entity and candidate-anchor task reported below; this is not a generic MTEB, MIEB, or MAEB SOTA claim.

SALT at 768d:

• image->text R@1: 0.1794
• text->image R@1: 0.1968

Speech holdout at 768d:

• audio->text R@1: 0.3870
• text->audio R@1: 0.3624

Entity/rejection:

• entity_key same/different entity AUC: 0.9953
• entity_key same-topic/different-entity rejection AUC: 0.9953
• entity_key same-entity/different-event rejection AUC: 0.8001
• weak-reference candidate R@1: 1.0000
• anchor-memory candidate R@1: 0.9647

Selected MTEB/MIEB/MAEB memory slice:

• 768d: 8 / 8 selected tasks complete, 0 exceptions
• 1536d: 8 / 8 selected tasks complete, 0 exceptions after rerun
• text: SprintDuplicateQuestions 0.9161 at 768d and 0.9323 at 1536d; STSBenchmark 0.7442 and 0.7535
• image-text: Flickr T2I R@1 0.1764 at 768d and 0.1864 at 1536d
• audio-text: Clotho R@1 0.0512 at 768d and 0.0514 at 1536d

Files

File	Purpose
ES-AIST-81M_q8_0.gguf	Higher-accuracy GGUF
ES-AIST-81M_q5_1.gguf	Smaller GGUF
manifest.json	Release manifest
parameter_breakdown.json	Exact parameter accounting
retrieval_768_1536_gt1030.json	Exact retrieval eval for the source checkpoint
entity_eval.json	Entity AUC eval
episode_aux_eval.json	Event/rejection eval
candidate_ranking_eval.json	Candidate-anchor ranking eval
signal_eval.json	Signal-level eval summary
M_SERIES_MEMORY_SLICE.md	Selected MTEB/MIEB/MAEB memory-slice report
es_aist_mseries_memory_slice_eventboost_summary.json	Machine-readable selected slice summary

Notes

• This is a preview quantization repo.
• SALT is held out from ES training and is included as a regression/generalization gate.
• The source model emits semantic and entity signals only; reference resolution remains engine-side.
• Full MTEB/MIEB/MAEB reporting is future work; the included slice is selected memory-relevant smoke coverage.