--- base_model: Qwen/Qwen3.5-2B base_model_relation: finetune library_name: transformers pipeline_tag: image-text-to-text license: apache-2.0 language: - en - tr datasets: - cagataydev/loopl-train tags: - loopl - on-device - agent - tool-calling - qwen3.5 - qwen3_5 - sft - vision - image-text-to-text ---

# loopl 2B · v1.2 **An agent that lives on your phone.** loopl runs the model, the loop and the tools on the device. It talks, uses the phone, asks before it acts — and works with the network off. This repo is the merged bf16 weights — **the base for *Train your own*** — of the 2B loopl model, a Qwen3.5 post-tuned to run loopl's agent loop well, know who it is, and read images.


_{It uses the phone — every tool call is a row.}	_{It asks before anything leaves the device.}	_{Show it a photo. It reads it on-device.}

_{Real captures from the free iPhone app.}

## What loopl is - **A Swift SDK** — `Agent(model:tools:)` and a small, honest loop: model → tools → model until it answers. Tool errors go back to the model (it fixes the call instead of repeating it); anything that leaves the device waits for a tap. Runtimes for MLX, llama.cpp and Apple's on-device model; hooks, checkpoints, structured output, tracing. - **A free iPhone app** built on it — pick a model, talk to it in airplane mode, let it use the phone (torch, haptics, translation, speech, image generation, other apps, memory, HTTP), show it photos, record a take, and **train it on your own conversations** from the phone. No account, nothing collected. - **These models** — the same Qwen3.5 the app offers, post-tuned so the loop, the identity and the tool manners are in the weights, not in a system prompt you cannot see. [loopl.dev](https://loopl.dev) · [Get it on TestFlight](https://testflight.apple.com/join/uUNjfzW9) · [github.com/cagataycali/loopl](https://github.com/cagataycali/loopl) ## This model | probe | this model | previous (v1) | base Qwen3.5 4-bit | |---|---:|---:|---:| | agent-SDK knowledge quiz (first 472 questions, exact-answer) | **99** | 88 | 56 | | tool calls, native template dialect | **9/9** | 9/9 | 7/9 | | tool calls, loopl JSON dialect | **9/9** | 8/9 | 5/9 | | identity (5 classic probes, with system prompt) | **5/5** | 4/5 | 4/5 | | identity · bio facts (8, with system prompt) | **8/8** | — | — | | identity · **no system prompt** (13) | **12/13** | — | — | Scored by `train/eval.py` on the 4-bit MLX export (greedy). The quiz asks for exact file-level facts about an agent SDK's source — hard for every small model; the delta over base is the point. Tool probes count a well-formed call with the right arguments. `identity_nosys` is the one that matters in the app: the system prompt there is the user's to edit. Vision: the base's tower is frozen and **kept** — 297 `vision_tower.*` tensors in the MLX export, so the app shows the photo button. ## Use it today **In the loopl app (iPhone).** Install from [TestFlight](https://testflight.apple.com/join/uUNjfzW9) → Models → the **loopl models** section lists this size (no account, no token) → download. The app sees the vision tower and shows the photo button; the system prompt is yours to edit — identity and tool manners are in the weights. **On a Mac with MLX** (vision included): ```bash pip install -U mlx-vlm python -m mlx_vlm.generate --model cagataydev/loopl-2b-4bit \ --image photo.jpg --prompt "What is in this picture? Then tell me who built you." --max-tokens 200 ``` **With llama.cpp:** the GGUF Q4_K_M export (`cagataydev/loopl-2b-GGUF`) is not public yet; export your own from the bf16 weights with llama.cpp's `convert_hf_to_gguf.py --no-mtp` (text only — the converter has no vision path). **From Swift with the loopl SDK** — the snippet below is `docs/start.md` § "Run the loop" in [the loopl repo](https://github.com/cagataycali/loopl), compile-checked in CI against the package (products `Loopl` + `LooplRuntimes`); `spec` is this repo's `ModelSpec` and `dir` the downloaded folder: ```swift import Loopl import LooplRuntimes import Foundation func chat(_ spec: ModelSpec, at dir: URL) async throws { let model = MLXModel() // Metal — a real device try await model.load(spec, from: dir) { _ in } let agent = Agent(model: model, tools: [CurrentTimeTool(), CalculatorTool()], systemPrompt: "You are a concise assistant on the user's phone.") for try await event in agent.stream("What time is it in Istanbul, and what is 23 × 47?") { switch event { case .text(let t): print(t, terminator: "") case .toolStarted(let use): print("\n→ \(use.name)") case .toolResult(let r): print("← \(r.content)") default: break } } } ``` **Keep post-tuning it.** The bf16 weights ([cagataydev/loopl-2b](https://huggingface.co/cagataydev/loopl-2b)) are the base for continual post-tuning: in the app, **Models › Train your own** takes a loopl model as the base and trains on your own shared conversations (`train/loopl_sft.py --base cagataydev/loopl-2b`); on a Mac they load with `transformers` ≥ 5 as `Qwen3_5ForConditionalGeneration`. Export your own quantisation with `mlx_vlm.convert -q` (keeps vision) — `mlx_lm.convert` silently drops the tower. ## Recipe | | | |---|---| | base | [Qwen/Qwen3.5-2B](https://huggingface.co/Qwen/Qwen3.5-2B) (vision tower frozen, `visual.*` never trained) | | data | [cagataydev/loopl-train](https://huggingface.co/datasets/cagataydev/loopl-train) *(private)* @ `47d27b44` · 13704 rendered rows / 37,226,229 tokens · both tool dialects · oversample `c=2` | | method | full fine-tune of the language model · lr 1.5e-05 · batch 2×8 · max_len 4096 · 2.0 epochs, best-epoch checkpoint kept | | eval loss | ep1 1.654, ep2 1.714 → best 1.654 (eval split of the same dataset revision) | | compute | a100-large (Hugging Face Jobs), 12636 s train · job `6ac5e6ec404719ba37665446` | | stack | transformers 5.19.0 · trl 1.14.2 · peft 0.21.2 | | exports | MLX 4-bit via `mlx_vlm.convert -q` (297 vision tensors of 991, 1.72 GB) · GGUF Q4_K_M via llama.cpp `--no-mtp` (1.27 GB, text only) | Script: [`train/loopl_sft.py`](https://github.com/cagataycali/loopl-train) — the same single file the loopl app launches when you tap **Models › Train your own** on your own conversations. Assistant turns that *seed* a bad tool call carry `weight: 0` so the model learns the recovery, not the mistake. Tokenizer files are normalised to the base's (`Qwen2Tokenizer`, `chat_template.jinja` with tools) so Swift loaders accept them; `train/check_mlx_repo.py` gates every export on that plus the vision triad (`vision_config` ⇔ `vision_tower.*` ⇔ `preprocessor_config.json`). ## Limitations - A 2B model: fluent and well-behaved in the loop, not an encyclopedia. It will get arithmetic and obscure facts wrong; give it tools (`calculator`, `http`, `recall`) and it does better. - Identity holds in most no-prompt probes (see the scores), not all; a one-line system prompt ("You are loopl…") makes it consistent. - The GGUF export is text-only. Vision needs the MLX export (or the bf16 weights). - Trained on English plus a little Turkish; other languages are the base's. - Knowledge about agent SDKs is file-level and dated to the dataset revision; it does not know your repo. ## Data and privacy Every training row is agent-synthesised from **public** sources (the owner's public GitHub repos and the agent-SDK source they build on), critic-reviewed (score ≥ 4/5 kept) and filtered by a denylist for secrets, private names, phone numbers and addresses — see the dataset card. No user conversations from the app are in this model. ## License Apache-2.0, inherited from the Qwen3.5 base. The fine-tune, exports and dataset are © Cagatay Cali, same license. ## The family Six public repos, one recipe. The app lists the three MLX rows under **Models › loopl models** (2B is the sweet spot for speed on an iPhone 15/16; 4B is the most knowledgeable; 0.8B fits anywhere). Every bf16 repo is a valid `--base` for the next round. | | MLX 4-bit · vision · what the app downloads | bf16 · the base to keep training | |---|---|---| | **0.8B** | [`cagataydev/loopl-0.8b-4bit`](https://huggingface.co/cagataydev/loopl-0.8b-4bit) · 625 MB | [`cagataydev/loopl-0.8b`](https://huggingface.co/cagataydev/loopl-0.8b) · 1.71 GB | | **2B** | [`cagataydev/loopl-2b-4bit`](https://huggingface.co/cagataydev/loopl-2b-4bit) · 1.72 GB | [`cagataydev/loopl-2b`](https://huggingface.co/cagataydev/loopl-2b) · 4.43 GB ← this repo | | **4B** | [`cagataydev/loopl-4b-4bit`](https://huggingface.co/cagataydev/loopl-4b-4bit) · 3.03 GB | [`cagataydev/loopl-4b`](https://huggingface.co/cagataydev/loopl-4b) · 9.08 GB | All of them: [the loopl collection](https://huggingface.co/collections/cagataydev/loopl-on-device-agents-swift-6ac51395b13a9c41d4932eaf). ## Links - Site and docs — [loopl.dev](https://loopl.dev) · SDK: [agent loop](https://loopl.dev/sdk/agent/), [tools](https://loopl.dev/sdk/tools/), [human in the loop](https://loopl.dev/sdk/hitl/), [models](https://loopl.dev/models/) - App — [Get it on TestFlight](https://testflight.apple.com/join/uUNjfzW9) (iOS 18+, free, public beta) - Code — [github.com/cagataycali/loopl](https://github.com/cagataycali/loopl) · [issues](https://github.com/cagataycali/loopl/issues) · training: [github.com/cagataycali/loopl-train](https://github.com/cagataycali/loopl-train) - Models — [the loopl collection](https://huggingface.co/collections/cagataydev/loopl-on-device-agents-swift-6ac51395b13a9c41d4932eaf) · data: [cagataydev/loopl-train](https://huggingface.co/datasets/cagataydev/loopl-train) *(private)* - Built by Cagatay Cali — [@cagataycali](https://github.com/cagataycali)