#!/usr/bin/env bash
# Download F3 (Q5_K_M / Q8_0 quantizations + quantized DFlash drafts).
# Does not compete with the GPU (network/disk only). Uses scripts/dl.py (resume+retries).
# Usage: setsid nohup bash scripts/download_f3.sh > /tmp/opencode/dl-f3.log 2>&1 &
set -uo pipefail
REPO="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
cd "$REPO"
source scripts/spec-env.sh

DL="python scripts/dl.py"
B="https://huggingface.co"
mkdir -p models

log() { echo "[$(date '+%F %T')] $*"; }

log "F3 downloads START"
# Quantized targets (F3): Qwen3-8B Q5/Q8, Gemma 4 12B Q5/Q8
$DL "$B/Qwen/Qwen3-8B-GGUF/resolve/main/Qwen3-8B-Q5_K_M.gguf" models/Qwen3-8B-Q5_K_M.gguf && log "qwen q5 OK"
$DL "$B/Qwen/Qwen3-8B-GGUF/resolve/main/Qwen3-8B-Q8_0.gguf" models/Qwen3-8B-Q8_0.gguf && log "qwen q8 OK"
$DL "$B/unsloth/gemma-4-12b-it-GGUF/resolve/main/gemma-4-12b-it-Q5_K_M.gguf" models/gemma-4-12b-it-Q5_K_M.gguf && log "gemma q5 OK"
$DL "$B/unsloth/gemma-4-12b-it-GGUF/resolve/main/gemma-4-12b-it-Q8_0.gguf" models/gemma-4-12b-it-Q8_0.gguf && log "gemma q8 OK"
# Quantized draft (G3): DFlash 12B Q4_K_M / Q8_0
$DL "$B/williamliao/gemma-4-12B-it-DFlash-GGUF/resolve/main/gemma-4-12B-it-DFlash-Q4_K_M.gguf" models/drafts/gemma-4-12B-it-DFlash-Q4_K_M.gguf && log "dflash q4 OK"
$DL "$B/williamliao/gemma-4-12B-it-DFlash-GGUF/resolve/main/gemma-4-12B-it-DFlash-Q8_0.gguf" models/drafts/gemma-4-12B-it-DFlash-Q8_0.gguf && log "dflash q8 OK"
log "F3 downloads DONE"