Total runs:
1.1M
Run Growth:
58.6K
Growth Rate:
5.18%
Sponsored by
Agentkit
litert-community / Bonsai-Image-ternary-4B
Model's Last Updated: September 05 2026
text-to-image
Introduction of Bonsai-Image-ternary-4B
Model Details of Bonsai-Image-ternary-4B
Bonsai Image 4B — LiteRT (.tflite)
PrismML Bonsai Image 4B
— a ternary-weight diffusion transformer built on the FLUX.2-klein-4B architecture — converted to LiteRT for on-device text-to-image. The whole pipeline runs in three fixed-shape
.tflite
graphs; the host does only tokenization, the FlowMatch-Euler loop, and latent unpatchify (
generate.py
in this repo, ~150 lines, no torch and no diffusers).
| graph | file | recipe | size |
|---|---|---|---|
| DiT (Flux2Transformer2DModel, 3.88 B) |
dit_int4b32.tflite
|
int4 block-32 | 2.11 GiB |
| text encoder (Qwen3-4B, top 9 layers pruned) |
textenc_int4.tflite
|
int4 block-128 DRQ | 1.68 GiB |
| text encoder — higher-fidelity variant |
textenc_int8_weightonly.tflite
|
int8 weight-only | 2.91 GiB |
| VAE decoder (AutoencoderKLFlux2) |
vae_dec_fp32.tflite
|
fp32 | 0.19 GiB |
Smallest working set: 3.97 GiB (int4 text encoder). Output is fixed at 512×512, 256 prompt tokens, 4 sampling steps by default (the model is step-distilled; more steps also work).
The ternary weights survive int4 exactly
The vendor ships the DiT with ternary weights (every block linear uses values {−scale, 0, +scale} per 128-group). In the int4 block-32 container those land as exactly {−7, 0, +7} — zero rounding decisions, verified in the converted artifact. The int4 file is a lossless container for this model's weights, at 2.11 GiB versus 14.4 GiB fp32.
Quality
Converting without quantizing anything is essentially exact (76.8 dB vs the PyTorch pipeline). With quantization, low PSNR against the PyTorch sample is sampler drift, not degradation : a small weight perturbation amplified through 4 sampling steps lands on a different draw from the same distribution. Across a 6-prompt stress set (legible text, a face, macro texture, a smooth gradient, a cluttered scene), reference-free sharpness stays flat across every recipe (Laplacian variance 1086 → 1069 at the lowest-PSNR variant), text stays legible, and faces stay correct in all columns. Pick a text encoder by size, not by fear: int4 if you want the smallest set, int8 weight-only if you want to track the reference sample more closely.
Run it
pip install ai-edge-litert numpy pillow transformers
python generate.py --model-dir . --prompt "a red fox sitting in fresh snow at sunrise" --seed 42 --out fox.png
Timings on an Apple-silicon Mac (CPU, 8 threads): DiT ~3.9 s/step, text encoder 2.6 s, VAE 1.3 s — ~19 s per 512×512 image at 4 steps.
On-device (iPhone 17 Pro, CPU via XNNPACK, 6 threads):
the 2.11 GiB DiT loads and prepares in
4.5 s; 13 s per DiT step, text encoder 1.8 s, VAE 3.1 s — **
64 s per 512×512 image** at 4 steps, ~2.9 GiB peak memory. Device output is bit-exact against the Mac run (every DiT step matches to the last float; 51.2 dB PSNR on the final PNG). One integration note: attach the XNNPACK delegate explicitly (with your thread count) when using the C API — without it the runtime falls back to reference kernels, which are orders of magnitude slower on this model.
Conversion
Converted with [litert-torch] via the recipe published at
hf-to-litertlm
(export scripts, quantization, and the two conversion gotchas: Flux2's float64 RoPE frequency table must be forced to float32 to legalize
tfl.pow
, and blockwise zero-scales in all-zero blocks need patching for XNNPACK). The text encoder is exported as a prompt embedder — the pipeline reads hidden states from layers (9, 18, 27) only, so the top 9 of 36 layers and the LM head are pruned for free.
License
Apache-2.0, following the upstream prism-ml/bonsai-image-ternary-4B release.
Runs of litert-community Bonsai-Image-ternary-4B on huggingface.co
3.0K
Total runs
32
24-hour runs
40
3-day runs
155
7-day runs
2.1K
30-day runs
More Information About Bonsai-Image-ternary-4B huggingface.co Model
Url of Bonsai-Image-ternary-4B
Bonsai-Image-ternary-4B huggingface.co Url
Provider of Bonsai-Image-ternary-4B huggingface.co
Other API from litert-community
Updated:August 31 2026
Total runs:
345.4K
Run Growth:
-37.4K
Growth Rate:
-11.05%
Total runs:
72.2K
Run Growth:
-1.8K
Growth Rate:
-2.52%
Total runs:
61.3K
Run Growth:
-7.9K
Growth Rate:
-12.91%
Total runs:
43.2K
Run Growth:
10.2K
Growth Rate:
23.62%
Total runs:
23.2K
Run Growth:
-857
Growth Rate:
-3.69%
Total runs:
14.3K
Run Growth:
-678
Growth Rate:
-4.73%
Total runs:
13.4K
Run Growth:
715
Growth Rate:
5.27%
Total runs:
13.2K
Run Growth:
2.3K
Growth Rate:
17.78%
Total runs:
7.1K
Run Growth:
-451
Growth Rate:
-6.38%
Total runs:
6.6K
Run Growth:
4.6K
Growth Rate:
70.75%
Total runs:
6.3K
Run Growth:
-1.5K
Growth Rate:
-24.14%
Total runs:
4.4K
Run Growth:
406
Growth Rate:
9.66%
Total runs:
3.9K
Run Growth:
3.5K
Growth Rate:
90.12%
Total runs:
3.6K
Run Growth:
681
Growth Rate:
19.14%
Total runs:
3.1K
Run Growth:
455
Growth Rate:
14.78%
Total runs:
2.5K
Run Growth:
2.3K
Growth Rate:
91.69%
Total runs:
2.4K
Run Growth:
936
Growth Rate:
39.11%
Total runs:
2.2K
Run Growth:
61
Growth Rate:
2.88%
Total runs:
2.2K
Run Growth:
517
Growth Rate:
23.85%
Total runs:
2.2K
Run Growth:
1.5K
Growth Rate:
67.82%
Total runs:
2.1K
Run Growth:
2.1K
Growth Rate:
100.00%
Total runs:
2.1K
Run Growth:
-96
Growth Rate:
-4.64%
Total runs:
2.0K
Run Growth:
891
Growth Rate:
43.78%
Total runs:
2.0K
Run Growth:
117
Growth Rate:
5.86%
Total runs:
1.9K
Run Growth:
7
Growth Rate:
0.37%
Total runs:
1.9K
Run Growth:
-2.0K
Growth Rate:
-105.57%
Total runs:
1.8K
Run Growth:
629
Growth Rate:
34.96%
Total runs:
1.7K
Run Growth:
787
Growth Rate:
46.87%
Total runs:
1.6K
Run Growth:
1.3K
Growth Rate:
80.33%
Total runs:
1.4K
Run Growth:
328
Growth Rate:
24.62%
Total runs:
1.3K
Run Growth:
984
Growth Rate:
73.21%
Total runs:
1.3K
Run Growth:
-107
Growth Rate:
-8.04%
Total runs:
1.3K
Run Growth:
-486
Growth Rate:
-38.79%
Total runs:
1.0K
Run Growth:
20
Growth Rate:
1.98%
Total runs:
968
Run Growth:
40
Growth Rate:
4.13%
Total runs:
956
Run Growth:
935
Growth Rate:
97.80%
Total runs:
894
Run Growth:
86
Growth Rate:
9.62%
Total runs:
752
Run Growth:
752
Growth Rate:
100.00%
Total runs:
735
Run Growth:
175
Growth Rate:
23.81%
Total runs:
716
Run Growth:
419
Growth Rate:
58.52%
Total runs:
645
Run Growth:
88
Growth Rate:
13.64%
Total runs:
628
Run Growth:
292
Growth Rate:
46.50%
Total runs:
613
Run Growth:
455
Growth Rate:
74.23%
Total runs:
602
Run Growth:
330
Growth Rate:
54.82%
Total runs:
560
Run Growth:
-656
Growth Rate:
-117.14%
Total runs:
491
Run Growth:
0
Growth Rate:
0.00%
Total runs:
486
Run Growth:
486
Growth Rate:
100.00%
Total runs:
485
Run Growth:
-47
Growth Rate:
-9.69%
Total runs:
464
Run Growth:
463
Growth Rate:
99.78%
Total runs:
462
Run Growth:
408
Growth Rate:
88.31%
Total runs:
453
Run Growth:
216
Growth Rate:
47.68%
Total runs:
426
Run Growth:
-50
Growth Rate:
-11.74%
Total runs:
425
Run Growth:
425
Growth Rate:
100.00%
Total runs:
418
Run Growth:
105
Growth Rate:
25.12%
Total runs:
414
Run Growth:
48
Growth Rate:
11.59%
Total runs:
410
Run Growth:
88
Growth Rate:
21.15%
Total runs:
406
Run Growth:
210
Growth Rate:
51.72%
Total runs:
392
Run Growth:
172
Growth Rate:
43.88%
Total runs:
369
Run Growth:
-116
Growth Rate:
-31.44%
Total runs:
366
Run Growth:
200
Growth Rate:
54.64%
Total runs:
364
Run Growth:
246
Growth Rate:
67.58%
Total runs:
357
Run Growth:
357
Growth Rate:
100.00%
Total runs:
355
Run Growth:
165
Growth Rate:
46.48%
Total runs:
340
Run Growth:
36
Growth Rate:
10.59%
Total runs:
314
Run Growth:
130
Growth Rate:
41.40%
Total runs:
304
Run Growth:
180
Growth Rate:
59.21%
Total runs:
297
Run Growth:
270
Growth Rate:
90.91%
Total runs:
271
Run Growth:
271
Growth Rate:
100.00%
Total runs:
267
Run Growth:
240
Growth Rate:
89.89%
Total runs:
260
Run Growth:
170
Growth Rate:
65.38%
Total runs:
258
Run Growth:
177
Growth Rate:
68.60%
Total runs:
253
Run Growth:
163
Growth Rate:
64.43%
Total runs:
251
Run Growth:
2
Growth Rate:
0.80%
Total runs:
249
Run Growth:
145
Growth Rate:
58.23%
Total runs:
246
Run Growth:
246
Growth Rate:
100.00%
Total runs:
241
Run Growth:
87
Growth Rate:
36.10%
Total runs:
231
Run Growth:
116
Growth Rate:
53.21%
Total runs:
226
Run Growth:
73
Growth Rate:
32.30%
Total runs:
224
Run Growth:
135
Growth Rate:
60.27%
Total runs:
220
Run Growth:
54
Growth Rate:
24.55%
Total runs:
216
Run Growth:
68
Growth Rate:
31.48%
Total runs:
212
Run Growth:
204
Growth Rate:
100.00%
Total runs:
212
Run Growth:
147
Growth Rate:
69.34%
Total runs:
203
Run Growth:
75
Growth Rate:
39.27%
Total runs:
202
Run Growth:
202
Growth Rate:
100.00%
Total runs:
183
Run Growth:
131
Growth Rate:
73.18%
Total runs:
174
Run Growth:
-21
Growth Rate:
-12.07%
Total runs:
173
Run Growth:
98
Growth Rate:
60.87%
Total runs:
172
Run Growth:
-12
Growth Rate:
-6.98%
Total runs:
171
Run Growth:
110
Growth Rate:
64.33%
Total runs:
169
Run Growth:
31
Growth Rate:
18.34%
Total runs:
169
Run Growth:
110
Growth Rate:
65.09%
Total runs:
165
Run Growth:
86
Growth Rate:
52.12%
Total runs:
164
Run Growth:
80
Growth Rate:
48.78%
Total runs:
161
Run Growth:
23
Growth Rate:
14.29%
Total runs:
159
Run Growth:
87
Growth Rate:
54.72%
Total runs:
158
Run Growth:
158
Growth Rate:
100.00%