sparrow-nano-general (v1.1)

A 2.2 MB general-purpose decision model. It answers yes/no (noul), pick-one (choice) and rate-on-a-scale (score) questions about a text state, with calibrated probabilities. It runs in 3.8 ms per decision on a $10 ESP32-S3 and in about 0.06 ms on a laptop core with the C runtime. It is the default model of the sparrow-decider pip package.

Part of Sparrow Decider, a family of decision models from 47 MB down to a few KB. Same question and answer shapes as Strands Decider.

Unofficial community project. Not an AWS or Strands product.

Model ID	sparrow-nano-general-v1.1
Tier	nano (static embeddings + MLP)
File	sparrow-nano-general-v1.1.sparrow (2.2 MB, Python) · sparrow-nano-general-v1.1.spb (2.53 MB packed SPB1 blob, C)
Input	text
Runs on	ESP32-S3, Raspberry Pi, laptop
Dependencies	numpy (Python) · C99 + libm, no malloc (C)
Also bundled in	pip install sparrow-decider (default model)

Intended use

• Tool and intent routing inside an agent, gating ("should I call the big model?"), urgency scores, short text decisions on edge devices.
• Any place where a decision must be made in milliseconds without a GPU, network or model server.

Out of scope: open-ended language far from its training data, yes/no judgement that needs reasoning (see Limitations), and being the only safety mechanism on a physical system. For decisions over sensor numbers, prefer sparrow-nano-robot.

How to use

Python

pip install sparrow-decider

from sparrow_decider import Decider

d = Decider.load()                       # the bundled sparrow-nano-general-v1.1
d.ask("User: please move my 3pm meeting with Sam to Friday", {
    "tool": {"type": "choice", "instructions": "Which tool should the agent call?",
             "criteria": {"weather": "get a weather forecast", "calendar": "read or change calendar events",
                          "email": "send or read email", "search": "search the web"}},
    "urgency": {"type": "score", "instructions": "How urgent is this?", "criteria": ["can wait", "today", "now"]},
})
# {"tool": {"type": "choice", "choice": "calendar", "probabilities": {...}, "confidence": ...},
#  "urgency": {"type": "score", "score": ..., ...}}

The exact file from this repo, pinned to a version:

from huggingface_hub import hf_hub_download
from sparrow_decider import Decider

path = hf_hub_download("sparrow-decider/sparrow-nano-general", "sparrow-nano-general-v1.1.sparrow", revision="v1.1")
d = Decider.load(path)

CLI and HTTP server: sparrow-decider ask "..." --choice "..." -o name=description, sparrow-decider serve --port 8000 (POST /v1/systemone).

C / ESP32-S3

sparrow-nano-general-v1.1.spb is the packed blob for the C99 runtime in c/sparrow (one .c, one .h, an ESP-IDF component). It is read in place from flash; no malloc.

#include "sparrow.h"
static sparrow_model m;
static sparrow_ctx ctx;   /* working memory; keep it static */

sparrow_load(&m, blob, blob_len);            /* blob = contents of the .spb, e.g. in flash */
sparrow_option opts[] = {{"weather", "get a weather forecast"}, {"calendar", "read or change calendar events"},
                         {"email", "send or read email"}, {"search", "search the web"}};
const char *state = "User: please move my 3pm meeting with Sam to Friday";
float conf;
int k = sparrow_choice(&m, &ctx, state, "Which tool should the agent call?", opts, 4, &conf);
float p_true = sparrow_noul(&m, &ctx, state, "Is the user upset?");

Build with -DSPARROW_MAX_DIM=128 to halve RAM for nano. On an ESP32-S3, follow the ESP32-S3 guide; to compile the model in as a C array use python c/tools/pack.py sparrow-nano-general-v1.1.sparrow -o model.spb --header nano_model.h --symbol nano_model.

Measured results

All numbers are from the project's bench/RESULTS.md and site/facts.json. Local runs, not leaderboard scores.

Speed

Where	Time per decision (p50)
ESP32-S3, 240 MHz (int8 SIMD)	3.8 ms (first on-chip build: 9.6 ms)
Laptop, 1 Apple M4 Pro core, C runtime	0.06 ms (63.5 µs)
Laptop, 1 core, numpy	0.21 ms (RobotBench) · 0.37 ms p50 / 4.64 ms p95 (JevBench)
Peak RSS, numpy, JevBench run	52.2 MB

Quality

Benchmark	sparrow-nano-general-v1.1	Strands Decider 2B (public)
JevBench public, 231 tasks (local v1.5 proxy)	122/231
MASSIVE 1.1 en-US, 600 held-out utterances: domain / intent	0.50 / 0.588	0.86 / 0.871
RobotBench v0, 55 hand-written robot decisions	32/55	40/55

C parity: on the host, the C runtime agrees with numpy on 286/286 argmaxes, max probability difference 1e-6. On the ESP32-S3, int8 builds agree with the host on 150-155 of 155 decisions; the few mismatches are rounding near ties.

Limitations

• Static embeddings pool words without order: negation and multi-step reasoning are weak.
• Open-ended language it never trained on is well behind the 2B (MASSIVE routing above).
• Yes/no judgement on JevBench is near chance at every Sparrow tier. A noul near 0.5 means "don't know"; treat it that way, or use a full decider or an LLM for those questions.
• Close calls between similar options can have low confidence; read confidence, not just choice.
• English only. Text normalisation is exact for Latin scripts.
• It suggests decisions. Never use it as the only safety stop on a physical system.

Training

• Init: potion-base-8M static embeddings, PCA-reduced to 128-d, vocabulary pruned to 16k pieces (every single character kept, so any word still tokenizes).
• Architecture: mean-pooled state / question / option vectors and an MLP (128 hidden) that scores every option against the state and the question.
• Data: a general decision mix plus generated robot decisions. Only tasksource rows marked for commercial use were used; sources keep their own licences.
• Teacher: distilled from a Strands Decider 4B checkpoint (targets 50% teacher, 50% gold) on about 430k decisions; per-kind temperature calibration.
• Compression: int8 matrices with a per-row scale.

Details: research report.

Licence and credits

• Licence: Apache-2.0 (see LICENSE and NOTICE in this repo).
• Base weights: derived from potion-base-8M (Model2Vec, MinishLab), released under the MIT licence. Its copyright notice and full MIT text are in LICENSE-potion-base-8M.txt in this repo and apply to the corresponding weights.
• Teacher and prompt format: Strands Decider (Apache-2.0).
• Evaluation only: Amazon MASSIVE 1.1 (CC BY 4.0), JevBench (MIT).

Unofficial community project. Not an AWS or Strands product.

Contact

• Email: vivek@sparrowdecider.com
• Issues: https://github.com/sparrow-decider/sparrow-decider/issues
• Discussions: https://github.com/sparrow-decider/sparrow-decider/discussions

Citation

@software{sparrow_decider,
  author  = {Raja P S, Vivek},
  title   = {Sparrow Decider: decision models from 47 MB down to a few KB},
  year    = {2026},
  version = {0.2.0},
  url     = {https://github.com/sparrow-decider/sparrow-decider}
}