Verdict 4 1.5B Quill
50-iteration fused LoRA SFT on Qwen2.5-Coder-1.5B-Instruct engineered for sub-4 GB free-RAM CPU laptops and offline background assistance. Not sold as a Cloud SKU.
Lineage, Provenance & Runtime Specification
- Upstream Checkpoint
- Qwen/Qwen2.5-Coder-1.5B-Instruct
- Upstream License
- Apache-2.0
- Architecture & Active Params
- qwen2 Dense · 1.5B dense
- Certified Context / Max Output
- 32,768 tok / 4,096 tok
- Recommended Hardware Tier
- < 12 GB RAM / CPU-only laptops (~3.52 GiB peak RSS at 32K f16 KV; 4 GB free RAM recommended)
- Weight Modification & Audit Proof
- Derived from Qwen/Qwen2.5-Coder-1.5B-Instruct (Apache-2.0), quantized Q5_K_M GGUF and 4-bit MLX.
Paired Empirical Evidence (Unmerged Comparators)
Artifact-Bound Benchmark Measurements
Compact Local Baseline & Memory Profile
Comparator: Single-arm release validation (release_manifest.json)
- BCB First-100
- 34/100
- CPU 32K Peak RSS
- 3.52 GiB
10/25 on BCB First-25 smoke
Idle 3.11 GiB at 32K f16 KV (0.88 GiB KV cache)
Methodology & Caveat: Designed for lightweight local autocomplete/edits on memory-constrained laptops, not sold as a Cloud chat SKU.
Live Basis P2 Burn & Effective Customer Rate Across Credit Packs
Certified Sampler Contract (§6.2)
Engine Presets & Cryptographic Checksums
- Local-only model ($0.00/tok); pair with vc/verdict/verdict-4-35b-toolterse-v5 on Cloud for heavy multi-file tasks.
llama.cpp (llama-server)
llama-server \
-m verdict-4-1500m-quill-Q5_K_M.gguf \
--ctx-size 32768 \
--temp 0.2 \
--repeat-penalty 1.0 \
--port 8080LM Studio (model.yaml)
# Install at ~/.lmstudio/hub/models/verdict/verdict-4-1500m-quill/model.yaml
model: verdict/verdict-4-1500m-quill
base: verdict/verdict-4-1500m-quill-gguf-q5-k-m/verdict-4-1500m-quill-Q5_K_M.gguf
config:
load:
fields:
- key: llm.load.contextLength
value: 32768
operation:
fields:
- key: llm.prediction.temperature
value: 0.2
- key: llm.prediction.repeatPenalty
value: 1.0
metadataOverrides:
domain: llm
architectures: [qwen2]
compatibilityTypes: [gguf]
paramsStrings: [1.5B]
minMemoryUsageBytes: 4000000000
contextLengths: [32768]
vision: false
reasoning: false
Verified Local Artifact Checksums
Binary Weights Staged — Gated on Phase 0b Ed25519 SigningSHA-256: 604f6903b69bbede5bacd8ac90975db380fd88992e00750c4e98d30785fff0ad
SHA-256: 452148f3ad66f7131501827d436475a3e727a8f75e1e1629b9c035c3e92c1dcf
weight_set_sha256: 874be5a0e1d1677fec75ade152364fd201951e42f35046232ce49571de8e8115
Known Tradeoffs & Guardrails
- Local-only tier; not hosted on api.verdictmodels.com.
- 1.5B parameter ceiling limits complex multi-step reasoning and unfamiliar library synthesis.