Theboukadida/laura-multilingual-qwen3-1.7b-GGUF overview
Laura multilingual application candidate Owner approved release. Private native acceptance and production fresh download verification passed. This Q4 K M model…
Runs locally from ~1.03 GB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| laura-multilingual-qwen3-1.7b-Q4_K_M.gguf | GGUF | Q4_K_M | 1.03 GB | Download |
Model Details
| Model ID | Theboukadida/laura-multilingual-qwen3-1.7b-GGUF |
|---|---|
| Author | Theboukadida |
| Pipeline | — |
| License | apache-2.0 |
| Base model | Qwen/Qwen3-1.7B |
| Last modified | 2026-10-19T12:29:15.000Z |
Model README
---
license: apache-2.0
base_model: Qwen/Qwen3-1.7B
language:
- de
- en
- fr
- ar
tags:
- gguf
- qwen3
- german-learning
- application-candidate
---
Laura multilingual application candidate
Owner-approved release. Private native acceptance and production fresh-download verification passed.
This Q4_K_M model supports the Laura offline Android German-learning app. Its training contains 6,567 reviewed grammar/fact and dialogue examples with English, French and Arabic explanations. It was trained from the original Qwen3-1.7B base for one epoch (411 updates, learning rate0.00002), with answer-only supervision and actual app-format conversation history. Base revision:70d244cc86ccca08cf5af4e1e306ecf908b1ad5e.
Intended use and limitations
Use with Laura's verified dictionary facts, grammar checks, response validation and reviewed guided-practice questions. The app's 23 guided scenarios are authored content; their passing dialogues do not demonstrate model-generated conversation quality.
The raw model is NOT accepted as a standalone German grammar reference. A separate36-item BF16 diagnostic produced17 passes,18 failures and one partial answer. It can praise incorrect German and invent explanations. The original base and earlier adapters also failed diagnostics. Do not treat this as a generally reliable teacher model.
Application verification
The grounded application pipeline passed review of394 turns across English, French and Arabic:319 original/grammar probes,24 dialogue turns,9 guided scenario starts 36 additional sentence checks and six unsupported-compound checks. Those36 sentences had no exact overlap in the training corpus at first evaluation; subsequent app fixes make them development checks. These numbers measure the combined app pipeline, NOT raw model accuracy.
Source verification:137 suites,3,167 tests passed; TypeScript passed. Private build432 passed82 final response checks across the three languages, including24 offline dialogue turns, plus9 navigation/cancellation memory samples. Build430 previously passed163 functional checks. Production build433 then passed a fresh public Wi-Fi download with matching on-phone SHA-256, background download progress,51 response rechecks across all three languages (including24 with networking disabled), and9 navigation/cancellation samples.89 targeted release checks and TypeScript also passed.
Runtime
Context2048; thinking disabled; temperature0.1, top_p1, top_k50. Use the app's production system prompt and validators. Native Android verification on the test phone kept focused PSS at about1.81–1.90GB and released hidden-chat memory to about0.25–0.314GB.
Bytes:1107408640
SHA-256:eed7ceebe4e41aa4eef1ac18f2dc6eaf7487f710a3a67f822318d0a1fab0b997
MD5:8ed37244d4bc8405a0db5576b2ef15d6
GGUF converter:llama.cpp33c923db1b3fec0d20818e32facb4914534c415c.
Large files are held locally and on the owner's private HP server. The source backup is separate. Public hosting was approved by the owner on 19 October 2026.
Run Theboukadida/laura-multilingual-qwen3-1.7b-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models