Model card: repo is public; link the hosted demo Space

#1
Files changed (1) hide show
  1. README.md +12 -10
README.md CHANGED
@@ -31,11 +31,12 @@ hybrid recurrence/attention backbone. Text, visual codes, audio codes, triangle
31
  meshes and tool-call proposals share one vocabulary and model. Frozen codecs
32
  convert between media and discrete tokens.
33
 
34
- **Release status: private, runnable research checkpoint release.** The default
35
  root weights are the completed curriculum-research checkpoint shown below; the
36
  preserved v5 comparison weights are in `variants/v5`. Matching source code,
37
  text tokenizers and pinned frozen visual/audio codecs are included. Training
38
- datasets, optimizer states and an inference service are not included.
 
39
  This card describes these two snapshots as of **21 October 2026**. Separate architecture
40
  experiments are still being evaluated and are excluded from these results.
41
 
@@ -225,13 +226,12 @@ false and was not relabeled as passed.
225
 
226
  This is a custom PyTorch checkpoint package, not a Transformers `AutoModel`.
227
  Use the included `quasar` implementation rather than
228
- `AutoModel.from_pretrained("Atlas-AI-research/Quasar")`. Authentication is needed
229
- while the repository is private. Preserve a working CUDA-compatible PyTorch
230
  installation; the measured environment used PyTorch 2.11.0+cu128. Install a suitable
231
  PyTorch build separately when needed, then install the remaining inference dependencies:
232
 
233
  ```bash
234
- hf auth login
235
  hf download Atlas-AI-research/Quasar --local-dir Quasar
236
  cd Quasar
237
  python -m pip install -r requirements-inference.txt
@@ -263,8 +263,10 @@ frames = generator.generate_video(
263
  ```
264
 
265
  These examples do not establish acceptable output for arbitrary prompts. GPU use
266
- must be exclusive with local training/evaluation workers. No inference server is
267
- deployed. About 1.96 GiB of model/package files include both 361 MB Quasar weight
 
 
268
  files and one shared copy of each frozen codec; training data and optimizer states
269
  are omitted. The image/code mappings and original trained weights are unchanged.
270
 
@@ -282,10 +284,10 @@ claim new semantic quality or replace the held-out results above.
282
 
283
  ## Licensing and provenance
284
 
285
- This private upload does not select a distribution license for Quasar's code or
286
- weights. No permissive license badge is implied by using PyTorch or permissively
287
  licensed components. Component notices remain separate and are included with
288
- their codec files; this upload does not make the repository public.
289
 
290
  The existing frozen Emu3 visual tokenizer is pinned at revision
291
  `c81f916ad371289e205310a7539255e8a9396488` under Apache-2.0. The local Descript DAC
 
31
  meshes and tool-call proposals share one vocabulary and model. Frozen codecs
32
  convert between media and discrete tokens.
33
 
34
+ **Release status: public, runnable research checkpoint release.** The default
35
  root weights are the completed curriculum-research checkpoint shown below; the
36
  preserved v5 comparison weights are in `variants/v5`. Matching source code,
37
  text tokenizers and pinned frozen visual/audio codecs are included. Training
38
+ datasets, optimizer states and an inference service are not included; a hosted demo runs in the
39
+ [Atlas AI Models Space](https://huggingface.co/spaces/AtlasAI-official/Atlas-AI-Models).
40
  This card describes these two snapshots as of **21 October 2026**. Separate architecture
41
  experiments are still being evaluated and are excluded from these results.
42
 
 
226
 
227
  This is a custom PyTorch checkpoint package, not a Transformers `AutoModel`.
228
  Use the included `quasar` implementation rather than
229
+ `AutoModel.from_pretrained("Atlas-AI-research/Quasar")`. The repository is public, so no
230
+ login is needed to download it. Preserve a working CUDA-compatible PyTorch
231
  installation; the measured environment used PyTorch 2.11.0+cu128. Install a suitable
232
  PyTorch build separately when needed, then install the remaining inference dependencies:
233
 
234
  ```bash
 
235
  hf download Atlas-AI-research/Quasar --local-dir Quasar
236
  cd Quasar
237
  python -m pip install -r requirements-inference.txt
 
263
  ```
264
 
265
  These examples do not establish acceptable output for arbitrary prompts. GPU use
266
+ must be exclusive with local training/evaluation workers. A public demo of the root
267
+ weights (chat, 64/128 px images, 8-frame video and image description) runs on ZeroGPU in the
268
+ [Atlas AI Models Space](https://huggingface.co/spaces/AtlasAI-official/Atlas-AI-Models).
269
+ About 1.96 GiB of model/package files include both 361 MB Quasar weight
270
  files and one shared copy of each frozen codec; training data and optimizer states
271
  are omitted. The image/code mappings and original trained weights are unchanged.
272
 
 
284
 
285
  ## Licensing and provenance
286
 
287
+ The repository is public, but this upload does not select a distribution license for Quasar's
288
+ code or weights. No permissive license badge is implied by using PyTorch or permissively
289
  licensed components. Component notices remain separate and are included with
290
+ their codec files.
291
 
292
  The existing frozen Emu3 visual tokenizer is pinned at revision
293
  `c81f916ad371289e205310a7539255e8a9396488` under Apache-2.0. The local Descript DAC