Project OpenTARS commited on
Commit
f5fac60
·
verified ·
1 Parent(s): ccfcebf

Update model notes and licenses for 5.3

Browse files
Files changed (1) hide show
  1. index.html +2 -2
index.html CHANGED
@@ -44,7 +44,7 @@ a.mail{color:var(--green)}
44
  <a class="card" href="https://huggingface.co/OpenTARS/Gargantua-Pi"><b>Gargantua Pi Gen 1</b><span>For small boards like the Raspberry Pi and the Orange Pi, in openTARS Pi.</span><em>0.3 GB</em></a>
45
  <a class="card" href="https://huggingface.co/OpenTARS/Murph"><b>Murph</b><span>A tiny request classifier that picks the kind of task in 0.2 ms, with no GPU.</span><em>4 MB</em></a>
46
  </div>
47
- <p class="note"><b>Built on Gemma 4:</b> Gargantua 2.5 (E4B) and Gargantua Lite Gen 3 (E2B) are published. Lite Gen 3 is trained for one epoch and 2.5 is an intermediate checkpoint (1,200 of 1,985 steps); training continues. Locating a button by position is weak in both, and audio is not tested yet.</p>
48
  <p class="note">The models are GGUF files with SHA-256 checksums. Every model card lists the tests we ran and the known limits. They are our own tests on one machine, not public benchmarks.</p>
49
 
50
  <h2>The app</h2>
@@ -55,7 +55,7 @@ a.mail{color:var(--green)}
55
  </div>
56
 
57
  <footer>
58
- The openTARS code is MIT. The Gargantua models are Apache-2.0, based on Qwen3.5 (Apache-2.0).
59
  Contact: <a class="mail" href="mailto:projectopentars@gmail.com">projectopentars@gmail.com</a>
60
  </footer>
61
  </div>
 
44
  <a class="card" href="https://huggingface.co/OpenTARS/Gargantua-Pi"><b>Gargantua Pi Gen 1</b><span>For small boards like the Raspberry Pi and the Orange Pi, in openTARS Pi.</span><em>0.3 GB</em></a>
45
  <a class="card" href="https://huggingface.co/OpenTARS/Murph"><b>Murph</b><span>A tiny request classifier that picks the kind of task in 0.2 ms, with no GPU.</span><em>4 MB</em></a>
46
  </div>
47
+ <p class="note"><b>Built on Gemma 4:</b> Gargantua 2.5 (E4B) and Gargantua Lite Gen 3 (E2B) power openTARS 5.3, with vision and audio.</p>
48
  <p class="note">The models are GGUF files with SHA-256 checksums. Every model card lists the tests we ran and the known limits. They are our own tests on one machine, not public benchmarks.</p>
49
 
50
  <h2>The app</h2>
 
55
  </div>
56
 
57
  <footer>
58
+ The openTARS code is MIT. Each model keeps the license of its base: Gemma 4 (Apache-2.0), Qwen2.5-Coder (Apache-2.0) and FunctionGemma (Gemma terms).
59
  Contact: <a class="mail" href="mailto:projectopentars@gmail.com">projectopentars@gmail.com</a>
60
  </footer>
61
  </div>