Fix bos/eos token ids and regenerate model.pte

Companion to the `main` fix for huggingface/optimum-executorch#28.

Sets `bos_token_id=1, eos_token_id=2` in `config.json` and `generation_config.json`, and regenerates `model.pte` from the corrected config so its `get_bos_id`/`get_eos_id` metadata match the tokenizer.

Verified with optimum-executorch main (44fa1d3), transformers 5.0.0rc1, executorch 1.3.1:
- current `executorch` revision: `ValueError: The tokenizer's bos_token_id=1 must be the same as the model's bos_token_id=0.`
- this PR: model reports bos=1, eos=[2]; `text_generation` runs.

Export command: `optimum-cli export executorch --model <fixed main> --task text-generation --recipe xnnpack`. Note: the new `model.pte` is 4.2 MB vs 8.4 MB for the current one, likely different export defaults from when it was first made; happy to re-export with specific flags if you'd prefer.

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+model.pte filter=lfs diff=lfs merge=lfs -text

config.json

@@ -5,8 +5,8 @@
   ],
   "attention_bias": false,
   "attention_dropout": 0.0,
-  "bos_token_id": 0,
-  "eos_token_id": 1,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 16,
   "initializer_range": 0.02,

generation_config.json

@@ -1,7 +1,7 @@
 {
   "_from_model_config": true,
-  "bos_token_id": 0,
-  "eos_token_id": 1,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
   "pad_token_id": -1,
   "transformers_version": "4.38.0.dev0"
 }

model.pte

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8e06bd2a8dfbd3fd53b1646cd1eec4876715cca6af9929f919eb8121366a04f9
-size 8382544
+oid sha256:1e8fa6d4a91268876210bdb81bfb32436ac65da97ceec131da5c0c2dc8661898
+size 4205952