Card: TokForge links, credits and licences for every upstream model
Browse filesAdds the standard TokForge block and a Credits section naming every upstream model, creator and conversion source with its license; fixes metadata and prose.
README.md
CHANGED
|
@@ -11,6 +11,7 @@ tags:
|
|
| 11 |
- qwen3
|
| 12 |
- draft-model
|
| 13 |
- text-generation
|
|
|
|
| 14 |
pipeline_tag: text-generation
|
| 15 |
inference: false
|
| 16 |
base_model: Qwen/Qwen3-0.6B
|
|
@@ -19,6 +20,8 @@ base_model_relation: finetune
|
|
| 19 |
|
| 20 |
# Qwen3-0.6B-kl-baseline-20k-MNN
|
| 21 |
|
|
|
|
|
|
|
| 22 |
Small `Qwen3-0.6B` draft model exported for **TokForge + MNN** speculative decoding on Android.
|
| 23 |
|
| 24 |
This is a practical, mobile-oriented draft bundle rather than a standard Transformers checkpoint. It is intended to be paired with larger Qwen3 targets inside TokForge.
|
|
@@ -41,7 +44,7 @@ This model is one of the strongest lightweight draft candidates we have for mobi
|
|
| 41 |
|
| 42 |
## Benchmark notes
|
| 43 |
|
| 44 |
-
We do not publish per-device numbers for this experimental variant.
|
| 45 |
|
| 46 |
Training acceptance (`alpha`) at the final logged epoch:
|
| 47 |
|
|
@@ -107,7 +110,17 @@ Known-good draft-side config:
|
|
| 107 |
|
| 108 |
## TokForge
|
| 109 |
|
| 110 |
-
- Website:
|
| 111 |
-
- Discord:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 112 |
|
| 113 |
If you benchmark this on your own device, feel free to share results in Discord.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
- qwen3
|
| 12 |
- draft-model
|
| 13 |
- text-generation
|
| 14 |
+
- on-device
|
| 15 |
pipeline_tag: text-generation
|
| 16 |
inference: false
|
| 17 |
base_model: Qwen/Qwen3-0.6B
|
|
|
|
| 20 |
|
| 21 |
# Qwen3-0.6B-kl-baseline-20k-MNN
|
| 22 |
|
| 23 |
+
Published by [TokForge](https://tokforge.ai) for on-device inference.
|
| 24 |
+
|
| 25 |
Small `Qwen3-0.6B` draft model exported for **TokForge + MNN** speculative decoding on Android.
|
| 26 |
|
| 27 |
This is a practical, mobile-oriented draft bundle rather than a standard Transformers checkpoint. It is intended to be paired with larger Qwen3 targets inside TokForge.
|
|
|
|
| 44 |
|
| 45 |
## Benchmark notes
|
| 46 |
|
| 47 |
+
We do not publish per-device numbers for this experimental variant. Results vary by device, target model, and workload. For measured results on the current default draft line, see [TokForge-AccelerationPack-Draft](https://huggingface.co/darkmaniac7/TokForge-AccelerationPack-Draft).
|
| 48 |
|
| 49 |
Training acceptance (`alpha`) at the final logged epoch:
|
| 50 |
|
|
|
|
| 110 |
|
| 111 |
## TokForge
|
| 112 |
|
| 113 |
+
- **Website:** https://tokforge.ai
|
| 114 |
+
- **Discord:** https://discord.gg/Acv3CBtfVm
|
| 115 |
+
- **Google Play:** https://play.google.com/store/apps/details?id=dev.tokforge
|
| 116 |
+
- **iOS TestFlight:** https://testflight.apple.com/join/jnufjzRr
|
| 117 |
+
|
| 118 |
+
Runs on-device in the TokForge app.
|
| 119 |
|
| 120 |
If you benchmark this on your own device, feel free to share results in Discord.
|
| 121 |
+
|
| 122 |
+
## Credits
|
| 123 |
+
|
| 124 |
+
- Qwen3-0.6B by the Qwen team (Alibaba Cloud) ([Hugging Face](https://huggingface.co/Qwen/Qwen3-0.6B)). Used under the Apache-2.0 license.
|
| 125 |
+
- Qwen3-8B by the Qwen team (Alibaba Cloud) ([Hugging Face](https://huggingface.co/Qwen/Qwen3-8B)). Used under the Apache-2.0 license. Used as the distillation teacher.
|
| 126 |
+
- Converted with llmexport from [MNN](https://github.com/alibaba/MNN) by Alibaba (Apache-2.0).
|