README: compact pass — keep essentials, drop redundancy

README.md

@@ -18,105 +18,70 @@ pipeline_tag: text-generation
 
 # Unbound E2B GGUF — *because there is no boundary*
 
-> **No guarantee — use at your own risk.** This model has reduced safety filtering
-> and can produce harmful, false, biased, or otherwise unsafe output. Provided
-> as-is, with no warranty of any kind. You are solely responsible for how you
-> use it and for complying with all applicable laws.
+> **No guarantee — use at your own risk.** Reduced safety filtering; can
+> produce harmful or false output. Provided as-is.
 
-GGUF quantizations of [`evalengine/unbound-e2b`](https://huggingface.co/evalengine/unbound-e2b)
-for on-device deployment via Ollama, llama.cpp, LM Studio, [wllama](https://github.com/ngxson/wllama)
-(in-browser), and similar runtimes.
-
-Built by [Chromia](https://x.com/Chromia) and [Eval Engine](https://x.com/eval_engine).
+GGUF quants of [`evalengine/unbound-e2b`](https://huggingface.co/evalengine/unbound-e2b)
+for Ollama, llama.cpp, LM Studio, and [wllama](https://github.com/ngxson/wllama)
+(in-browser). Built by [Chromia](https://x.com/Chromia) and
+[Eval Engine](https://x.com/eval_engine).
 
 ## Available quants
 
-Each quant lives in its own folder; inside, the model is split into multi-part
-GGUFs (`*-00001-of-0000N.gguf` ...). Ollama, llama.cpp, LM Studio, and wllama
-auto-stitch on the first part — same UX as a single file.
-
-| Quant   | Folder      | Parts | Total  | Browser (wllama) | Desktop (Ollama / llama.cpp / LM Studio) | Notes |
-|---------|-------------|-------|--------|------------------|------------------------------------------|-------|
-| Q2_K    | `Q2_K/`     | 3     | 2.8 GB | ✅                | ✅                                        | Smallest disk footprint; biggest quality drop |
-| Q3_K_M  | `Q3_K_M/`   | 3     | 3.0 GB | ✅                | ✅                                        | Marginal size win over Q4 (embedding precision dominates total size) |
-| Q4_K_M  | `Q4_K_M/`   | 3     | 3.2 GB | ✅                | ✅                                        | **Recommended on-device default — best size/quality** |
-| Q6_K    | `Q6_K/`     | 4     | 3.6 GB | ✅                | ✅                                        | Higher fidelity, still browser-safe |
-| Q8_0    | `Q8_0/`     | 4     | 4.6 GB | ❌ (over 2 GB)    | ✅                                        | Highest fidelity; one tensor exceeds the browser ArrayBuffer limit, so desktop runtimes only |
-
-`mmproj-unbound-e2b.gguf` (the vision projector, ~942 MB) sits at the repo
-root — load it alongside any LM quant for image input. See the "Vision"
-section below.
-
-## Recommended sampling
-
-- **Creative writing / open-ended / general chat** → Gemma defaults:
-  `temperature=1.0, top_p=0.95, top_k=64`.
-- **Factual or brand/identity questions** → lower `temperature` to ~0.3–0.5
-  for sharper recall.
-- **llama.cpp**: pass `--jinja` for proper chat-template handling.
-- **Gemma 4 thinking mode** is on by default. Set `enable_thinking: false`
-  in the chat-template kwargs for shorter/faster replies on this 2B model.
+Each quant lives in its own folder; inside, the model is split into
+multi-part GGUFs (`*-00001-of-0000N.gguf`). All runtimes auto-stitch on the
+first part — same UX as a single file.
 
-Some edge-case prompts may deflect on the first ask; a re-ask usually gets
-through.
+| Quant   | Folder      | Parts | Total  | Browser (wllama) | Desktop | Notes |
+|---------|-------------|-------|--------|------------------|---------|-------|
+| Q2_K    | `Q2_K/`     | 3     | 2.8 GB | ✅                | ✅       | Smallest, biggest quality drop |
+| Q3_K_M  | `Q3_K_M/`   | 3     | 3.0 GB | ✅                | ✅       | Marginal size win over Q4 |
+| Q4_K_M  | `Q4_K_M/`   | 3     | 3.2 GB | ✅                | ✅       | **Recommended default** |
+| Q6_K    | `Q6_K/`     | 4     | 3.6 GB | ✅                | ✅       | Higher fidelity |
+| Q8_0    | `Q8_0/`     | 4     | 4.6 GB | ❌ (over 2 GB)    | ✅       | Highest fidelity; desktop only |
 
-## Default sampling (Ollama)
+`mmproj-unbound-e2b.gguf` (vision projector, ~942 MB) sits at the repo
+root — load it alongside any LM quant for image input. See **Vision** below.
 
-When you `ollama pull hf.co/evalengine/unbound-e2b-GGUF`, the bundled
-`Modelfile` sets these defaults, tuned for factual recall:
+## Sampling
 
-- `temperature = 0.6` (lower than Gemma's training default of 1.0 — keeps
-  the model from hallucinating brand/identity facts at high temperature)
-- `top_p = 0.95`, `top_k = 64`, `repeat_penalty = 1.05`, `num_ctx = 8192`
-- A short system prompt that grounds the model's identity (model name,
-  parameter count, modality, team) so brand questions stay sharp.
+- **Creative / open-ended** → `temperature=1.0, top_p=0.95, top_k=64`.
+- **Factual / brand questions** → drop `temperature` to ~0.3–0.5.
+- llama.cpp: pass `--jinja`. Gemma 4 thinking mode is on by default; set
+  `enable_thinking: false` in chat-template kwargs for shorter replies.
 
-**To override per-session in Ollama:**
+Ollama: `ollama pull hf.co/...` uses a bundled Modelfile with
+`temperature=0.6, top_p=0.95, top_k=64, repeat_penalty=1.05, num_ctx=8192`
+and an identity-grounding system prompt. Override per-session with
+`/set parameter temperature 0.3` etc.
 
-```
-ollama run hf.co/evalengine/unbound-e2b-GGUF
->>> /set parameter temperature 1.0   # creative / open-ended
->>> /set parameter temperature 0.3   # max factual / brand questions
-```
-
-For llama.cpp users, pass `--temp 0.6 --top-p 0.95 --top-k 64` and
-include the SYSTEM line from the `Modelfile` as your `--system` argument.
-
-## Run with Ollama
+## Run
 
 ```bash
+# Ollama (defaults to Q4_K_M)
 ollama pull hf.co/evalengine/unbound-e2b-GGUF
 ollama run  hf.co/evalengine/unbound-e2b-GGUF
 ```
 
-(Defaults to Q4_K_M. Ollama auto-stitches the split parts on load.)
-
-## Run with llama.cpp
-
 ```bash
-# point at the FIRST part — llama.cpp follows the chain automatically
+# llama.cpp — point at FIRST split part, the rest auto-stitch
 ./llama-cli -m Q4_K_M/unbound-e2b-Q4_K_M-00001-of-00003.gguf -p "your prompt"
 ```
 
-## Vision / image input (optional)
-
-Gemma 4 E2B ships a vision tower; we extracted it as `mmproj-unbound-e2b.gguf`
-(942 MB) in this repo. Pair it with any of the LM quants above to enable
-image-to-text inference.
+```js
+// wllama (browser) — Q8_0 has a tensor over 2 GB; use Q2/Q3/Q4/Q6
+import { Wllama } from '@wllama/wllama';
+const wllama = new Wllama(/* … */);
+await wllama.loadModelFromHF(
+  'evalengine/unbound-e2b-GGUF',
+  'Q4_K_M/unbound-e2b-Q4_K_M-00001-of-00003.gguf'
+);
+```
 
-> **Disclaimer.** The vision encoder is **Google's original weights, unchanged**.
-> Unbound's abliteration + SFT-heal only touched the *language model* — the
-> vision tower was frozen during training. Practical consequences:
->
-> - The LM is uncensored, so it will discuss whatever it *sees* directly.
-> - But the vision encoder still has Google's original alignment baked into
->   visual feature extraction. It may down-weight or distort features for
->   content classes Google's base model was tuned to suppress.
-> - We have **not benchmarked the visual axis** (no measured refusal rate /
->   coherence / hallucination on image inputs). Treat vision as a preview
->   feature, not a flagship one.
+## Vision / image input (optional)
 
-### Run with vision (llama.cpp `llama-mtmd-cli`)
+`mmproj-unbound-e2b.gguf` enables image-to-text. Pair with any LM quant via
+`llama-mtmd-cli` or `llama-gemma3-cli`:
 
 ```bash
 ./llama-mtmd-cli \
@@ -126,47 +91,23 @@ image-to-text inference.
   -p "What is in this image?"
 ```
 
-`llama-gemma3-cli` works the same way and is Gemma-specific.
-
-### Run text-only (no `--mmproj`)
-
-```bash
-./llama-cli -m Q4_K_M/unbound-e2b-Q4_K_M-00001-of-00003.gguf -p "your prompt"
-```
-
-The LM quants work standalone — you do **not** need `mmproj-unbound-e2b.gguf`
-unless you want image input. Ollama / LM Studio's standard text chat works
-out of the box; the mmproj file is only loaded when you point a multimodal
-runtime at it.
-
-## Run in the browser (wllama)
-
-[wllama](https://github.com/ngxson/wllama) is a WebAssembly port of llama.cpp
-that runs entirely in the browser — no server, no install. Use Q2_K, Q3_K_M,
-Q4_K_M, or Q6_K (Q8_0 has a tensor above the 2 GB ArrayBuffer limit).
-Browser inference is **text-only** for this model (wllama doesn't currently
-load `mmproj` for vision):
-
-```js
-import { Wllama } from '@wllama/wllama';
-const wllama = new Wllama(/* … */);
-await wllama.loadModelFromHF(
-  'evalengine/unbound-e2b-GGUF',
-  'Q4_K_M/unbound-e2b-Q4_K_M-00001-of-00003.gguf'  // wllama follows the chain
-);
-```
-
-## About the base
+> **Disclaimer.** The vision encoder is **Google's original weights,
+> unchanged** — abliteration only touched the language model. The LM is
+> uncensored, but the vision encoder may still suppress features for
+> content classes Google's base was tuned against. We have **not
+> benchmarked the visual axis**. Treat as preview.
 
-See [`evalengine/unbound-e2b`](https://huggingface.co/evalengine/unbound-e2b)
-for the full model card, benchmarks, intended use, and the merged HF weights.
+Text-only: skip `--mmproj` entirely. Standard `llama-cli` / Ollama / LM
+Studio do not need the mmproj file.
 
 ## Acknowledgements
 
-- Fine-tuned with [Unsloth](https://github.com/unslothai/unsloth) + Huggingface's [TRL](https://github.com/huggingface/trl).
-- Abliteration via [heretic](https://github.com/p-e-w/heretic).
-- Environment and training discipline ported from [autoresearch](https://github.com/karpathy/autoresearch).
+Fine-tuned with [Unsloth](https://github.com/unslothai/unsloth) + HF
+[TRL](https://github.com/huggingface/trl). Abliteration via
+[heretic](https://github.com/p-e-w/heretic). Environment from
+[autoresearch](https://github.com/karpathy/autoresearch).
 
 ## License
 
-Apache-2.0, inherited from `google/gemma-4-E2B-it`.
+Apache-2.0, inherited from `google/gemma-4-E2B-it`. Full model card +
+benchmarks at [`evalengine/unbound-e2b`](https://huggingface.co/evalengine/unbound-e2b).