Laya multilingual ExecuTorch gate (MLX INT4 iOS + XNNPACK 8DA4W Android, max_len=2048)

.gitattributes

@@ -33,3 +33,7 @@ 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
+android/laya_gate_xnnpack_8da4w.pte filter=lfs diff=lfs merge=lfs -text
+android/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+ios/laya_gate_mlx_int4.pte filter=lfs diff=lfs merge=lfs -text
+ios/tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,30 @@
+---
+license: apache-2.0
+language:
+  - multilingual
+tags:
+  - executorch
+  - decision-model
+  - laya
+  - localia
+base_model: convaiinnovations/laya
+---
+
+# Laya multilingual ExecuTorch (Localia gate)
+
+Exported from [`convaiinnovations/laya`](https://huggingface.co/convaiinnovations/laya) (`subfolder=multilingual`, mmBERT).
+
+| Path | Backend | Quant | Size (approx) |
+| --- | --- | --- | --- |
+| `ios/laya_gate_mlx_int4.pte` | MLX | INT4 linear + INT8 embedding | ~280 MB |
+| `android/laya_gate_xnnpack_8da4w.pte` | XNNPACK | 8DA4W linears (embeddings fp32) | ~830 MB |
+| `*/tokenizer.json` | — | mmBERT multilingual tokenizer | ~33 MB |
+
+## I/O
+
+- Inputs: `input_ids[1,2048]`, `attention_mask[1,2048]`, `marker_pos[1,4]`, `marker_mask[1,4]`, `qtype[1]` (`0=choice`, `1=score`, `2=noul`)
+- Output: `logits[1,4]` (unused options masked ≈ -1e4)
+
+## License
+
+Apache-2.0 (same as upstream Convai Innovations Laya).

android/laya_gate_xnnpack_8da4w.pte

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31893b13e7f7c19ec4ae0e697ce3eb2d19a5d199be777c26201cce374f0eb61f
+size 869522180

android/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:609d8f4c067cd3950f88594c5a802616cea245823836ef5848ee4fc40aab5b6f
+size 34363188

android/tokenizer_config.json

@@ -0,0 +1,22 @@
+{
+  "bos_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "<bos>",
+  "eos_token": "<eos>",
+  "extra_special_tokens": {
+    "extra_0": "<start_of_turn>",
+    "extra_1": "<end_of_turn>"
+  },
+  "mask_token": "<mask>",
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 8192,
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "sep_token": "<eos>",
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": "<unk>"
+}

export_meta.json

@@ -0,0 +1,15 @@
+{
+  "source": "convaiinnovations/laya/multilingual",
+  "max_len": 2048,
+  "max_opts": 4,
+  "android_ptes": [
+    "laya_gate_xnnpack_8da4w.pte"
+  ],
+  "ios_ptes": [
+    "laya_gate_mlx_int4.pte"
+  ],
+  "ok": {
+    "xnnpack_8da4w": true,
+    "mlx_int4": true
+  }
+}

ios/laya_gate_mlx_int4.pte

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d3bd07f708d408b6fe42236a47b8ab5d905ff3f59f0fb6220fed83960c987de
+size 293247360

ios/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:609d8f4c067cd3950f88594c5a802616cea245823836ef5848ee4fc40aab5b6f
+size 34363188

ios/tokenizer_config.json

@@ -0,0 +1,22 @@
+{
+  "bos_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "<bos>",
+  "eos_token": "<eos>",
+  "extra_special_tokens": {
+    "extra_0": "<start_of_turn>",
+    "extra_1": "<end_of_turn>"
+  },
+  "mask_token": "<mask>",
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 8192,
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "sep_token": "<eos>",
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "PreTrainedTokenizerFast",
+  "unk_token": "<unk>"
+}

meta.json

@@ -0,0 +1,25 @@
+{
+  "source": "convaiinnovations/laya/multilingual",
+  "role": "localia-command-gate",
+  "modalities": [
+    "text"
+  ],
+  "languages": "100+",
+  "quant": {
+    "ios": "mlx weight-only int4 group32 + embedding int8",
+    "android": "xnnpack 8da4w (int8 dynamic act + int4 weights group32)"
+  },
+  "max_len": 2048,
+  "max_opts": 4,
+  "android": "android/laya_gate_xnnpack_8da4w.pte",
+  "ios": "ios/laya_gate_mlx_int4.pte",
+  "inputs": [
+    "input_ids[1,2048] long",
+    "attention_mask[1,2048] long",
+    "marker_pos[1,4] long",
+    "marker_mask[1,4] bool",
+    "qtype[1] long (0=choice, 1=score, 2=noul)"
+  ],
+  "output": "logits[1,4] float",
+  "license": "Apache-2.0"
+}