multilingual-e5-small → Core ML (fp16, enumerated shapes, pooling inside)

.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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

MultilingualE5Small.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:121f9988a394565994f5aa26cd8a86a7d32b76e0e5b986ecee07070e5b2f8b27
+size 283336

MultilingualE5Small.mlpackage/Data/com.apple.CoreML/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff0331d7b3e44efd122d3592caf82a09e1d88a0bbd458fc8a24223ebb14a3ba7
+size 235028736

MultilingualE5Small.mlpackage/Manifest.json

@@ -0,0 +1,18 @@
+{
+    "fileFormatVersion": "1.0.0",
+    "itemInfoEntries": {
+        "E8077791-CA4B-40BC-AB6D-7AD2E8D812F4": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Specification",
+            "name": "model.mlmodel",
+            "path": "com.apple.CoreML/model.mlmodel"
+        },
+        "F2378018-85B9-4DA1-B8A7-7AB3C0E11498": {
+            "author": "com.apple.CoreML",
+            "description": "CoreML Model Weights",
+            "name": "weights",
+            "path": "com.apple.CoreML/weights"
+        }
+    },
+    "rootModelIdentifier": "E8077791-CA4B-40BC-AB6D-7AD2E8D812F4"
+}

README.md

@@ -0,0 +1,49 @@
+---
+license: mit
+base_model: intfloat/multilingual-e5-small
+language:
+- ru
+- en
+- multilingual
+tags:
+- coreml
+- sentence-embeddings
+- e5
+library_name: coreml
+pipeline_tag: feature-extraction
+---
+
+# multilingual-e5-small · Core ML
+
+[intfloat/multilingual-e5-small](https://huggingface.co/intfloat/multilingual-e5-small)
+converted to a Core ML **ML Program** (fp16) for on-device semantic search in
+the DND Master app (iPad / iPhone / Mac).
+
+## What is inside
+
+| File | Purpose |
+|---|---|
+| `MultilingualE5Small.mlpackage` | Core ML model. Inputs `input_ids`, `attention_mask` (`int32`, shape `[1, L]`, L ∈ {16, 32, 64, 128, 256, 512}); output `embedding` `[1, 384]` — masked mean pooling and L2 normalisation are inside the graph. |
+| `tokenizer.json`, `tokenizer_config.json`, `special_tokens_map.json`, `config.json` | XLM-R tokenizer files from the upstream repo, unchanged. |
+
+Pad token id is `1` (`<pad>`), attention mask is 1 for real tokens. Use the
+e5 prefixes: `query: ` for queries, `passage: ` for documents.
+
+## Conversion notes
+
+- coremltools 9.0, torch 2.7.0, transformers 4.46.3, `minimum_deployment_target = iOS18`.
+- Enumerated input shapes instead of a `RangeDim`: with a flexible range Core ML
+  returned NaN (“Data-dependent shapes were disabled”).
+- The additive attention mask was replaced by `-1e4` before tracing: the
+  default `finfo(float32).min` overflows to `-inf` in fp16 and the softmax
+  produces NaN.
+- Parity with the PyTorch reference: cosine 1.000000 on test sentences; parity
+  with an MLX fp16 embedding of a 12 270-chunk Russian corpus: min cosine 0.9997.
+- Recommended compute units: `.cpuAndNeuralEngine`. The GPU path is ~25× slower
+  for this graph on Apple silicon.
+
+The conversion script lives in the app repository (`BestiaryEmbedder/convert-e5-coreml.py`).
+
+## License
+
+The weights are derived from `intfloat/multilingual-e5-small` (MIT).

config.json

@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "tmp/",
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 384,
+  "initializer_range": 0.02,
+  "intermediate_size": 1536,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "tokenizer_class": "XLMRobertaTokenizer",
+  "torch_dtype": "float32",
+  "transformers_version": "4.29.0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 250037
+}

special_tokens_map.json

@@ -0,0 +1,9 @@
+{
+  "bos_token": "<s>",
+  "cls_token": "<s>",
+  "eos_token": "</s>",
+  "mask_token": "<mask>",
+  "pad_token": "<pad>",
+  "sep_token": "</s>",
+  "unk_token": "<unk>"
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b44a9d7b51c3c62626640cda0e2c2f70fdacdc25bbbd68038369d14ebdf4c39
+size 17082730

tokenizer_config.json

@@ -0,0 +1,20 @@
+{
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "<s>",
+  "eos_token": "</s>",
+  "mask_token": {
+    "__type": "AddedToken",
+    "content": "<mask>",
+    "lstrip": true,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "model_max_length": 512,
+  "pad_token": "<pad>",
+  "sep_token": "</s>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "XLMRobertaTokenizer",
+  "unk_token": "<unk>"
+}