Instructions to use dcostenco/prism-coder-1.7b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dcostenco/prism-coder-1.7b with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="dcostenco/prism-coder-1.7b",
	filename="prism-aac-1b7-q4km.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use dcostenco/prism-coder-1.7b with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf dcostenco/prism-coder-1.7b:Q8_0
# Run inference directly in the terminal:
llama-cli -hf dcostenco/prism-coder-1.7b:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf dcostenco/prism-coder-1.7b:Q8_0
# Run inference directly in the terminal:
llama-cli -hf dcostenco/prism-coder-1.7b:Q8_0

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf dcostenco/prism-coder-1.7b:Q8_0
# Run inference directly in the terminal:
./llama-cli -hf dcostenco/prism-coder-1.7b:Q8_0

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf dcostenco/prism-coder-1.7b:Q8_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf dcostenco/prism-coder-1.7b:Q8_0

Use Docker

docker model run hf.co/dcostenco/prism-coder-1.7b:Q8_0

LM Studio
Jan
Ollama
How to use dcostenco/prism-coder-1.7b with Ollama:
```
ollama run hf.co/dcostenco/prism-coder-1.7b:Q8_0
```

Unsloth Studio new

How to use dcostenco/prism-coder-1.7b with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for dcostenco/prism-coder-1.7b to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for dcostenco/prism-coder-1.7b to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for dcostenco/prism-coder-1.7b to start chatting

Pi new

How to use dcostenco/prism-coder-1.7b with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf dcostenco/prism-coder-1.7b:Q8_0

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "dcostenco/prism-coder-1.7b:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use dcostenco/prism-coder-1.7b with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf dcostenco/prism-coder-1.7b:Q8_0

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default dcostenco/prism-coder-1.7b:Q8_0

Run Hermes

hermes

Docker Model Runner
How to use dcostenco/prism-coder-1.7b with Docker Model Runner:
```
docker model run hf.co/dcostenco/prism-coder-1.7b:Q8_0
```

Lemonade

How to use dcostenco/prism-coder-1.7b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull dcostenco/prism-coder-1.7b:Q8_0

Run and chat with the model

lemonade run user.prism-coder-1.7b-Q8_0

List all available models

lemonade list

dcostenco commited on 13 days ago

Commit

6ee010d

verified ·

1 Parent(s): cb6eebe

Update README: v41 (96.1% BFCL) with per-category table

Browse files

Files changed (1) hide show

README.md +61 -35

README.md CHANGED Viewed

@@ -3,56 +3,82 @@ language: en
 license: apache-2.0
 base_model: Qwen/Qwen3-1.7B
 tags:
-  - tool-calling
-  - routing
-  - aac
   - gguf
-  - mlx
 ---
-# prism-coder:1b7 — AAC Tool Router (1.7B)
-Fine-tuned from **Qwen3-1.7B** for deterministic tool routing in the [Prism AAC](https://github.com/dcostenco/prism-aac) system.
-**BFCL accuracy: 100%** on 100-case × 3 seeds routing benchmark (v36 corpus).
-## What it does
-Routes user messages to one of 6 tools or plain text with zero hallucination:
-| Tool | Trigger |
-|------|---------|
-| `session_load_context` | Load/fetch context for project X |
-| `session_save_ledger` | Note / jot down / log / remember |
-| `session_save_handoff` | Handoff to next agent / pass on |
-| `session_compact_ledger` | Compact/archive/trim the ledger |
-| `session_search_memory` | What did we discuss / recall session |
-| `knowledge_search` | What do I know / stored notes |
-| *(plain text)* | AAC phrases, math, facts, translation, time |
-## Deployment
-**iOS / edge** — runs on-device via llama.cpp (1.0 GB, Q4_K_M):
-```bash
-ollama run dcostenco/prism-coder:1b7
-```
 ## Files
-| File | Size | Format |
-|------|------|--------|
-| `prism-coder-1b7-v36-q4km.gguf` | 1.0 GB | Q4_K_M GGUF (recommended) |
-| `prism-aac-1b7-q4km.gguf` | 1.0 GB | Q4_K_M GGUF (legacy name) |
-## Training
-- **Base**: Qwen3-1.7B
-- **Method**: MLX LoRA fine-tuning (mlx_lm.lora)
-- **Dataset**: v36_1b7 routing corpus (414 examples, 6-tool system prompt)
-- **Hardware**: Apple Silicon (M-series), ~4GB RAM
-- **Eval**: BFCL 100-case benchmark × 3 seeds → **100%**
-## System prompt
-Uses the 13-rule routing system prompt. See [Prism AAC](https://github.com/dcostenco/prism-aac) for the canonical prompt used in training and inference.

 license: apache-2.0
 base_model: Qwen/Qwen3-1.7B
 tags:
+  - tool-routing
+  - function-calling
+  - prism-aac
+  - qwen3
   - gguf
 ---
+# prism-coder:1b7 — Tool Routing Model (Ultra-Compact / iOS Tier)
+Fine-tuned Qwen3-1.7B for 6-tool routing in the [Prism AAC](https://github.com/dcostenco/prism-aac) system.
+Primary deployment: **on-device iOS inference** via llama.cpp (1.1 GB GGUF, Q4_K_M).
+## BFCL Routing Benchmark — v41 (Current)
+**Mean: 96.1%** (3-seed average, seeds 2027/2028/2029, 102 cases each)
+| Category | Description | Accuracy |
+|----------|-------------|:--------:|
+| aac | AAC phrase requests → plain text | 100% |
+| cmpct | Ledger compaction | 83% |
+| edge | Multi-step / compound requests | 83% |
+| hand | Agent handoff / relay | 100% |
+| info | General facts → plain text | 100% |
+| irrel | Irrelevant / live queries → plain text | 90% |
+| know | Knowledge base search | 100% |
+| load | Session context loading | 89% |
+| pred | Factual / knowledge queries → plain text | 100% |
+| save | Session ledger save | 100% |
+| smem | Session memory search | 100% |
+| tran | Translation requests → plain text | 100% |
+Eval: Ollama inference, temperature=0, Qwen3 thinking suppressed (`<think>\n\n</think>`), num_predict=160.
+Gate: ≥90% = deploy.
+## Version History
+| Version | BFCL | Notes |
+|---------|------|-------|
+| v41 | 96.1% | Current — routing corpus v41 |
+| v40 | ~95% | Routing corpus v40 |
+| v39 | ~94% | Routing corpus v39 |
+| v36 | 100% | Previous — routing corpus v36 (small eval set) |
+## Tools
+The model routes between exactly 6 tools:
+1. `session_load_context` — load/fetch/resume project context
+2. `session_save_ledger` — note/log/remember/record progress
+3. `session_save_handoff` — handoff/relay to next agent/session
+4. `session_compact_ledger` — compact/archive/shrink ledger
+5. `session_search_memory` — recall past sessions/conversations
+6. `knowledge_search` — search stored notes/knowledge base
 ## Files
+| File | Size | Use |
+|------|------|-----|
+| `prism-coder-1b7-v41-q4km.gguf` | 1.1 GB | Ollama / desktop |
+| `prism-aac-1b7-q4km.gguf` | 1.1 GB | iOS app download |
+## Cascade Role
+Ultra-compact iOS/edge tier. Desktop cascade: **1.7B → 8B → 14B → 32B → cloud Claude**.
+1.7B handles offline on-device routing where memory is tightly constrained (< 2 GB available).
+## Usage (Ollama)
+```bash
+ollama run dcostenco/prism-coder:1b7
+```
+## Training
+- **Base**: `Qwen/Qwen3-1.7B` (fp16, 1.7B params)
+- **Framework**: MLX-LM LoRA (rank=8, scale=20, 4 layers)
+- **Data**: v41 routing corpus
+- **Merge**: Direct safetensors manipulation (delta = scale/rank × B^T A^T)
+- **Peak memory**: ~4 GB (M-series Mac)