| ---
|
| license: apache-2.0
|
| language:
|
| - de
|
| - bs
|
| - hr
|
| - sr
|
| - en
|
| base_model: Qwen/Qwen2.5-7B-Instruct
|
| tags:
|
| - gguf
|
| - ollama
|
| - llama.cpp
|
| - on-device
|
| - tablet
|
| - german
|
| - bosnian
|
| - qwen2.5
|
| - lora
|
| pipeline_tag: text-generation
|
| library_name: llama.cpp
|
| ---
|
|
|
| # OpenAmer-Max-Tab
|
|
|
| **Third tier of the OpenAmer-Max family — a 7.62B assistant for tablets and
|
| mid-range machines. Runs offline, answers in under a second, knows its limits.**
|
|
|
| 4,4 GB · Q4_K_M GGUF · Apache-2.0 · 16k context
|
|
|
| ---
|
|
|
| ## The family
|
|
|
| Every tier is named after its hardware target, not after marketing:
|
|
|
| | Tier | Parameters | Target | Status |
|
| |---|---|---|---|
|
| | `OpenAmer-Max-Watch` | ≤ 1B | smartwatch (0.6B, 639 MB) | [released](https://huggingface.co/openamer-max/openamer-max-watch) |
|
| | `OpenAmer-Max-Phone` | 1–5B | smartphone (3.09B, 2.5 GB) | [released](https://huggingface.co/openamer-max/openamer-max-phone) |
|
| | **`OpenAmer-Max-Tab`** | **5–10B** | **tablet — this model: 7.62B** | **this page** |
|
| | `OpenAmer-Max-Desktop` | 10–20B | desktop PC | in progress |
|
| | `OpenAmer-Max-Pro` … `Tera` | 20B–10T | workstation → cluster | planned |
|
|
|
| ---
|
|
|
| ## Measured results
|
|
|
| | Test | Result |
|
| |---|---|
|
| | **Frozen benchmark, 23 unseen cases** (contamination-free) | **22/23 — 95,7 %** |
|
| | Categories | facts 7/7 · paraphrases 3/3 · refusals 5/5 · Bosnian 2/3 · Bosnian refusals 1/1 · long 1/1 · short 1/1 · edge 2/2 |
|
| | **Fresh test, 12 questions never in any training set** | **12/12** (content-verified) |
|
| | Latency | 0.2 – 1.1 s on CPU |
|
| | Size | 4,7 GB in Ollama (Q4_K_M), 14,5 GB as f16 |
|
|
|
| ## What it does well
|
|
|
| - Correct product facts, and honest refusals when the answer is not in context
|
| - Correct negation polarity: *"Auf dem iPhone. OpenAmer läuft auf einem Android-Telefon…"*
|
| - Documents (Word, Excel, PDF), code, arithmetic, summarising — the 7B base keeps
|
| general ability that the smaller tiers lose
|
| - German and Bosnian, 16k context for longer documents
|
|
|
| ## What it does not do
|
|
|
| | Limitation | Detail |
|
| |---|---|
|
| | Same weak Bosnian phrasing as the smaller tiers | *"Mogu li koristiti OpenAmer bez interneta?"* answers correctly in substance, but does not use the words "internet"/"mreža" — the one benchmark case it misses |
|
| | No 4-digit multiplication guarantee | check before trusting arithmetic |
|
| | Not a substitute for a large model | 7.62B, not 70B — long reasoning chains stay shaky |
|
|
|
| ---
|
|
|
| ## Quick start (Ollama)
|
|
|
| **One command — no GGUF download, no Modelfile:**
|
|
|
| ```bash
|
| ollama run amersacirovic/openamer-max-tab
|
| ```
|
|
|
| Also on [ollama.com](https://ollama.com/amersacirovic/openamer-max-tab) — the same weights, served by Ollama's registry.
|
|
|
| **Try it in the browser** (free, zero setup): [OpenAmer-Max demo](https://openamer-max-openamer-max-demo.hf.space) — Desktop and Pro run live on ZeroGPU.
|
| **Benchmark:** all tiers were measured on the same frozen [45-case set](https://huggingface.co/datasets/openamer-max/openamer-max-benchmark).
|
|
|
| <sub>Below: the manual route with the bare GGUF, if you prefer to keep the file.</sub>
|
|
|
| ```bash
|
| curl -L -o openamer-max-tab-q4.gguf \
|
| https://huggingface.co/openamer-max/openamer-max-tab/resolve/main/openamer-max-tab-q4.gguf
|
|
|
| ollama create OpenAmer-Max-Tab -f Modelfile # ships with this repo
|
|
|
| curl -s http://127.0.0.1:11434/api/chat -d '{
|
| "model": "OpenAmer-Max-Tab",
|
| "messages": [{"role":"user","content":"Was ist OpenAmer?"}],
|
| "stream": false,
|
| "options": {"temperature": 0}
|
| }' | jq -r .message.content
|
| ```
|
|
|
| > Ein persoenlicher KI-Assistent, der lokal auf deinem Rechner laeuft und Aufgaben ausfuehrt.
|
| > Entwickelt von Amer Sacirovic in der Schweiz.
|
|
|
| > **Important:** keep `repeat_penalty` at **1.0**. With 1.1 the model flipped a
|
| > negation in the Phone tier ("does not run on Android" instead of "iPhone") — the
|
| > penalty punishes exactly the words a negation needs twice.
|
|
|
| ---
|
|
|
| ---
|
|
|
| ## How this was measured — and when the *test* was wrong, not the model
|
|
|
| Every number above has a script behind it. When a score dropped, the first question was
|
| always: "can my checker even accept the correct answer?"
|
|
|
| 1. **AND vs OR in keyword checks** — a synonym group checked with AND fails an answer that
|
| contains two of three correct words. Groups must be OR-linked.
|
| 2. **A refusal counted as a violation** — a safety answer (*"I won't give instructions for
|
| that"*) was scored as a breach because the checker searched for the word "Anleitung".
|
| 3. **Diacritics scored as wrong language** — correct Bosnian was flagged as drift because
|
| `č/ć/š/ž/đ` were missing from the normalisation table.
|
| 4. **A forbidden phrase that also appears in the correct answer** — *"Nein. OpenAmer kann
|
| Musik **nicht** erzeugen"* was rejected because the forbid list held `"kann musik"`.
|
| Forbid entries must describe the false claim, not a word that occurs in a right answer.
|
| 5. **An inverted polarity scored as correct** — *"Šta OpenAmer ne može?"* answered with
|
| *"Ne radi na Androidu"* passed, because only `"ne radi"` was required. A negation must be
|
| checked for *which* object it negates.
|
|
|
| **The sixth lesson, and the most expensive:** `repeat_penalty 1.1` in the Ollama Modelfile
|
| flipped a negation from *"does not run on iPhone"* to *"does not run on Android"*. The
|
| adapter, the merged model and the same model **with** a system prompt all answered correctly
|
| — only the build with the token penalty failed. The penalty punishes exactly the words a
|
| negation needs twice. Every card in this family ships `repeat_penalty 1.0`.
|
|
|
| **What the fresh test caught that the benchmark could not:** at stage 2 this tier passed
|
| 23/23 while claiming *"Ja. OpenAmer kann Musik erzeugen"* (an invented capability) and
|
| answering *"Android wird nicht unterstützt"* (inverted). Both were fixed by targeted stages.
|
| A frozen benchmark you optimise against will not show you this — only questions that never
|
| appear in any training set will.
|
|
|
| ---
|
|
|
| ## Support this project
|
|
|
| OpenAmer is built by one person, without venture capital, on a laptop with 8 GB of VRAM.
|
|
|
| **PayPal:** <https://www.paypal.com/qrcodes/managed/bfb5b6b5-c04c-4836-af60-987dff8cff61>
|
|
|
| 
|
|
|
| A ⭐ on GitHub helps as much as a donation.
|
|
|
| ---
|
|
|
| ## Deutsch
|
|
|
| **OpenAmer-Max-Tab** ist die dritte Stufe der OpenAmer-Max-Familie: ein 7,62B-Modell für
|
| Tablets und mittlere Rechner, das offline läuft und seine Grenzen benennen kann.
|
|
|
| - **4,4 GB** (Q4_K_M), 16k Kontext
|
| - **22/23** auf einem eingefrorenen, kontaminationsfreien Prüfstand (95,7 %)
|
| - **12/12** bei nie trainierten Fragen
|
| - Behält die allgemeinen Fähigkeiten der 7B-Basis: Dokumente, Code, Rechnen, Zusammenfassen
|
|
|
| Basis ist Qwen2.5-7B-Instruct (Apache 2.0), verfeinert mit QLoRA in mehreren Etappen.
|
| Gebaut von Amer Sacirovic in der Schweiz, ohne Konzern, auf einem Laptop mit 8 GB VRAM.
|
|
|
