Fix gptoss_from_pretrained to correctly load HuggingFace weights (#4)

* Fix gptoss_from_pretrained to correctly load HuggingFace weights

- Update gptoss_normalize_config to map HF config keys (num_local_experts,
  num_experts_per_tok, nested rope_scaling) to internal names
- Rewrite gptoss_hf_weights_remap to:
  - Use underscore suffix (_blocks/_scales) for MXFP4 weight detection
  - Remap HF parameter names to model parameter names
  - Concatenate separate q/k/v projections into combined qkv tensors

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* Add dotty to Imports

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* ++

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

Author: dfalbel

DESCRIPTION

@@ -11,14 +11,15 @@ Description: A collection of minimal implementations of deep learning
 License: MIT + file LICENSE
 Encoding: UTF-8
 Roxygen: list(markdown = TRUE)
-RoxygenNote: 7.3.2
+RoxygenNote: 7.3.3
 Suggests: 
     testthat (>= 3.0.0)
 Depends:
     R (>= 4.1.0)
 Config/testthat/edition: 3
-Imports: 
+Imports:
     cli,
+    dotty,
     hfhub,
     purrr,
     rlang,

NAMESPACE

@@ -12,6 +12,9 @@ export(gptbigcode_from_pretrained)
 export(gptneox)
 export(gptneox_from_config)
 export(gptneox_from_pretrained)
+export(gptoss)
+export(gptoss_from_config)
+export(gptoss_from_pretrained)
 export(hf_state_dict)
 export(llama)
 export(llama_from_config)