wbit5 commited on
Commit
64e4100
Β·
verified Β·
1 Parent(s): c7886eb

publish from LatentPortPort 7f79ac0

Browse files
Files changed (1) hide show
  1. README.md +9 -9
README.md CHANGED
@@ -14,8 +14,8 @@ pinned: false
14
  **Open tools and research for building software with AI agents** β€”
15
  disciplined workflows, safety, token efficiency, and reproducible model research.
16
 
17
- <a href="https://huggingface.co/collections/agentic-underground/prattlenot-6af7ac5e95e0fc80de563d61"><img alt="Collection: PrattleNot" src="https://img.shields.io/badge/collection-PrattleNot-6366f1"></a>
18
- <a href="https://agentic-underground.github.io/PrattleNot/"><img alt="Live demo in your browser" src="https://img.shields.io/badge/demo-runs%20in%20your%20browser-a855f7"></a>
19
  <a href="https://github.com/agentic-underground"><img alt="Source on GitHub" src="https://img.shields.io/badge/source-GitHub-24292f?logo=github"></a>
20
  <a href="https://huggingface.co/papers/2609.25053"><img alt="Builds on LatentPort by Simon Villani (arXiv 2609.25053)" src="https://img.shields.io/badge/builds%20on-LatentPort%20by%20Simon%20Villani-b31b1b"></a>
21
 
@@ -27,9 +27,9 @@ disciplined workflows, safety, token efficiency, and reproducible model research
27
  > prompt included, and rules set in a system prompt are broken more often bridged (measured on six pairs).
28
  > Do not rely on system-prompt instructions as guardrails for a bridged model; enforce policy outside it.
29
 
30
- ## On the Hub: the PrattleNot family
31
 
32
- **PrattleNot** reproduces [**LatentPort**](https://huggingface.co/papers/2609.25053), Simon Villani's method for
33
  handing a conversation from one language model to another *through its memory*. A small model reads
34
  the prompt; a fitted **bridge** translates its key/value cache into the memory a larger model would
35
  have built; the larger model answers from it β€” **without reading the prompt itself**, so its first
@@ -37,7 +37,7 @@ token arrives sooner.
37
 
38
  What we publish here are the **bridges**: the fitted weights that make that handoff work between a
39
  specific pair of models. Everything is gathered in the
40
- **[PrattleNot collection β†’](https://huggingface.co/collections/agentic-underground/prattlenot-6af7ac5e95e0fc80de563d61)**
41
 
42
  ### ✨ Featured β€” [`smollm2-135m-to-360m-bridge`](https://huggingface.co/agentic-underground/smollm2-135m-to-360m-bridge)
43
 
@@ -51,7 +51,7 @@ shapes, bridged with the paper's **one-token handoff** and its learned correctio
51
  | 🎯 **Held-out questions** | **6 of 7** correct β€” as many as the 360M reading everything |
52
  | πŸ” **Fidelity (NCR)** | **0.930** β€” the paper's corrected bridge on its own pair: 0.918 |
53
 
54
- ▢️ **[Try it in your browser](https://agentic-underground.github.io/PrattleNot/)** β€” both models and the
55
  bridge download once and run locally. Turn the bridge on, ask anything, and watch the counters show
56
  who read what.
57
 
@@ -60,7 +60,7 @@ who read what.
60
  Before carrying the method to new models, we reproduced LatentPort's own **4B β†’ 27B** pair on two
61
  RTX 3090s with upstream's timing code and statistic: **1.36Γ—** sooner at 128 prompt tokens, rising to
62
  **2.94Γ—** at ~2K and **3.39Γ—** at ~16K. The full tables, every run, and a step-by-step guide to
63
- repeat them are in the [PrattleNot repository](https://github.com/agentic-underground/PrattleNot).
64
 
65
  ### πŸ“„ The paper's own pair β€” [`qwen3.5-4b-to-9b-bridge`](https://huggingface.co/agentic-underground/qwen3.5-4b-to-9b-bridge)
66
 
@@ -239,7 +239,7 @@ and measured.
239
  what a published result measured.
240
  - 🧩 **Our own work only** β€” we host what we fitted. Base models stay with their publishers and are
241
  linked, never mirrored.
242
- - πŸ™ **Attributed** β€” PrattleNot is a derivative work of [LatentPort](https://latentport.com/) by
243
  Simon Villani ([source](https://github.com/LatentPort/latentport), Apache-2.0). Our bridges are
244
  Apache-2.0, except those fitted from Gemma, which are Gemma Model Derivatives, gated and distributed
245
  under the [Gemma Terms of Use](https://ai.google.dev/gemma/terms), and those fitted from Llama, which
@@ -253,4 +253,4 @@ Code plugins, a static auditor for agent skills, a token-aware scheduler), live
253
  **[GitHub β†’ agentic-underground](https://github.com/agentic-underground)**.
254
 
255
  Questions and corrections are welcome: open a discussion on the model concerned, or an
256
- [issue on PrattleNot](https://github.com/agentic-underground/PrattleNot/issues).
 
14
  **Open tools and research for building software with AI agents** β€”
15
  disciplined workflows, safety, token efficiency, and reproducible model research.
16
 
17
+ <a href="https://huggingface.co/collections/agentic-underground/prattle-not-6af7ac5e95e0fc80de563d61"><img alt="Collection: prattle_not" src="https://img.shields.io/badge/collection-prattle__not-6366f1"></a>
18
+ <a href="https://agentic-underground.github.io/prattle_not/"><img alt="Live demo in your browser" src="https://img.shields.io/badge/demo-runs%20in%20your%20browser-a855f7"></a>
19
  <a href="https://github.com/agentic-underground"><img alt="Source on GitHub" src="https://img.shields.io/badge/source-GitHub-24292f?logo=github"></a>
20
  <a href="https://huggingface.co/papers/2609.25053"><img alt="Builds on LatentPort by Simon Villani (arXiv 2609.25053)" src="https://img.shields.io/badge/builds%20on-LatentPort%20by%20Simon%20Villani-b31b1b"></a>
21
 
 
27
  > prompt included, and rules set in a system prompt are broken more often bridged (measured on six pairs).
28
  > Do not rely on system-prompt instructions as guardrails for a bridged model; enforce policy outside it.
29
 
30
+ ## On the Hub: the prattle_not family
31
 
32
+ **prattle_not** reproduces [**LatentPort**](https://huggingface.co/papers/2609.25053), Simon Villani's method for
33
  handing a conversation from one language model to another *through its memory*. A small model reads
34
  the prompt; a fitted **bridge** translates its key/value cache into the memory a larger model would
35
  have built; the larger model answers from it β€” **without reading the prompt itself**, so its first
 
37
 
38
  What we publish here are the **bridges**: the fitted weights that make that handoff work between a
39
  specific pair of models. Everything is gathered in the
40
+ **[prattle_not collection β†’](https://huggingface.co/collections/agentic-underground/prattle-not-6af7ac5e95e0fc80de563d61)**
41
 
42
  ### ✨ Featured β€” [`smollm2-135m-to-360m-bridge`](https://huggingface.co/agentic-underground/smollm2-135m-to-360m-bridge)
43
 
 
51
  | 🎯 **Held-out questions** | **6 of 7** correct β€” as many as the 360M reading everything |
52
  | πŸ” **Fidelity (NCR)** | **0.930** β€” the paper's corrected bridge on its own pair: 0.918 |
53
 
54
+ ▢️ **[Try it in your browser](https://agentic-underground.github.io/prattle_not/)** β€” both models and the
55
  bridge download once and run locally. Turn the bridge on, ask anything, and watch the counters show
56
  who read what.
57
 
 
60
  Before carrying the method to new models, we reproduced LatentPort's own **4B β†’ 27B** pair on two
61
  RTX 3090s with upstream's timing code and statistic: **1.36Γ—** sooner at 128 prompt tokens, rising to
62
  **2.94Γ—** at ~2K and **3.39Γ—** at ~16K. The full tables, every run, and a step-by-step guide to
63
+ repeat them are in the [prattle_not repository](https://github.com/agentic-underground/prattle_not).
64
 
65
  ### πŸ“„ The paper's own pair β€” [`qwen3.5-4b-to-9b-bridge`](https://huggingface.co/agentic-underground/qwen3.5-4b-to-9b-bridge)
66
 
 
239
  what a published result measured.
240
  - 🧩 **Our own work only** β€” we host what we fitted. Base models stay with their publishers and are
241
  linked, never mirrored.
242
+ - πŸ™ **Attributed** β€” prattle_not is a derivative work of [LatentPort](https://latentport.com/) by
243
  Simon Villani ([source](https://github.com/LatentPort/latentport), Apache-2.0). Our bridges are
244
  Apache-2.0, except those fitted from Gemma, which are Gemma Model Derivatives, gated and distributed
245
  under the [Gemma Terms of Use](https://ai.google.dev/gemma/terms), and those fitted from Llama, which
 
253
  **[GitHub β†’ agentic-underground](https://github.com/agentic-underground)**.
254
 
255
  Questions and corrections are welcome: open a discussion on the model concerned, or an
256
+ [issue on prattle_not](https://github.com/agentic-underground/prattle_not/issues).