Mirror of mlboydaisuke/Qwen3-Reranker-0.6B-CoreAI

.gitattributes

@@ -33,3 +33,5 @@ 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
+qwen3-reranker-0.6b_float16_s512_static.aimodel/main.mlirb filter=lfs diff=lfs merge=lfs -text
+tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,107 @@
+---
+license: apache-2.0
+base_model: Qwen/Qwen3-Reranker-0.6B
+tags:
+  - coreai
+  - text-ranking
+  - reranker
+  - apple-silicon
+  - on-device
+language:
+  - multilingual
+pipeline_tag: text-ranking
+---
+
+> **Mirror** of [`mlboydaisuke/Qwen3-Reranker-0.6B-CoreAI`](https://huggingface.co/mlboydaisuke/Qwen3-Reranker-0.6B-CoreAI) — the canonical repo ([CoreAI Model Zoo](https://github.com/john-rocky/coreai-model-zoo)). Updates land there first.
+
+
+# Qwen3-Reranker-0.6B — Core AI export
+
+[Qwen/Qwen3-Reranker-0.6B](https://huggingface.co/Qwen/Qwen3-Reranker-0.6B) as a single static
+Core AI graph for macOS 27 / iOS 27. The **cross-encoder** that closes the on-device RAG loop —
+embed (with [Qwen3-Embedding-0.6B-CoreAI](https://huggingface.co/mlboydaisuke/Qwen3-Embedding-0.6B-CoreAI))
+→ **rerank** → generate, all local and private.
+
+A cross-encoder reads one `query + document` sequence and asks the LM a yes/no question; the
+relevance score is the softmax weight on **"yes"** vs **"no"** at the final token. So it keeps the
+LM head (the embedder drops it), but it's still a plain `.aimodel` run via `AIModel.run` — one
+forward, no generation. The scoring tail (gather last token → head on that one position → 2-way
+softmax) is baked in-graph.
+
+## Graph contract
+
+| | name | shape | dtype |
+|---|---|---|---|
+| input | `input_ids` | [1, 512] | int32 (right-padded; pad id 151643) |
+| input | `attention_mask` | [1, 512] | int32 (1 = real, 0 = padding) |
+| output | `probs` | [1, 2] | fp16, `softmax([no, yes])` — **relevance = `probs[0,1]` = P(yes)** |
+
+## Host recipe
+
+Format the pair exactly like the upstream model card, then right-pad to 512:
+
+```python
+import coreai.runtime as rt, numpy as np
+from transformers import AutoTokenizer
+
+tok = AutoTokenizer.from_pretrained("tokenizer")
+PREFIX = ("<|im_start|>system\nJudge whether the Document meets the requirements based on the "
+          "Query and the Instruct provided. Note that the answer can only be \"yes\" or "
+          "\"no\".<|im_end|>\n<|im_start|>user\n")
+SUFFIX = "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n"
+INSTR  = "Given a web search query, retrieve relevant passages that answer the query"
+
+m = await rt.AIModel.load("qwen3-reranker-0.6b_float16_s512_static.aimodel",
+        rt.SpecializationOptions.from_preferred_compute_unit_kind(rt.ComputeUnitKind.gpu()))
+fn = m.load_function("main")
+
+def score(query, doc, S=512):
+    body = f"<Instruct>: {INSTR}\n<Query>: {query}\n<Document>: {doc}"
+    ids = (tok.encode(PREFIX, add_special_tokens=False)
+           + tok.encode(body, add_special_tokens=False)
+           + tok.encode(SUFFIX, add_special_tokens=False))
+    n = len(ids); ids = ids + [151643] * (S - n)
+    mask = [1] * n + [0] * (S - n)
+    res = await fn({"input_ids": rt.NDArray(np.asarray([ids], np.int32)),
+                    "attention_mask": rt.NDArray(np.asarray([mask], np.int32))})
+    return float(res["probs"].numpy()[0, 1])   # P(yes) = relevance; sort candidates by this
+```
+
+The instruction is swappable per task (the model is instruction-aware). Right-pad is equivalent to
+the upstream left-pad + `logits[:, -1]` (the graph reads the true last token from the mask).
+
+### Swift — [CoreAIKit](https://github.com/john-rocky/coreai-kit)
+
+Downloads this repo on first use and formats the pair in-process:
+
+```swift
+import CoreAIKitEmbeddings
+
+let reranker = try await Reranker(model: .qwen3Reranker0_6B)
+let ranked = try await reranker.rerank(
+    query: "What is the capital of Japan?",
+    documents: ["Tokyo is the capital of Japan.", "Python is a programming language."])
+// ranked[0].document is most relevant; ranked[i].score is P(yes) in [0, 1]
+```
+
+## Bundle layout
+
+```
+qwen3-reranker-0.6b_float16_s512_static.aimodel   (~1.1 GB, fp16)
+tokenizer/                                          (HF tokenizer files)
+reference.json                                      (pairs, scores, prompt scaffolding)
+```
+
+## Parity
+
+Precision **fp16**. Verified against the official `AutoModelForCausalLM` scoring (fp32): the
+in-graph wrapper reproduces P(yes) **exactly** (|Δ| = 0.00000 over 6 relevant/irrelevant pairs),
+relevant pairs 0.98–1.00 vs irrelevant ≈ 0.0000, ranking preserved. On the Core AI GPU delegate
+the `.aimodel` matches the torch reference within **|Δ| < 0.0005** end-to-end. Measured **45.7 ms
+per pair-score** on an M4 Max GPU (512 grid).
+
+## License
+
+Apache-2.0 (upstream model and code are Apache-2.0). Conversion script:
+[`conversion/export_qwen3_reranker.py`](https://github.com/john-rocky/coreai-model-zoo/blob/main/conversion/export_qwen3_reranker.py)
+in the coreai-model-zoo.

qwen3-reranker-0.6b_float16_s512_static.aimodel/main.hash

@@ -0,0 +1 @@
+&��o�i_e��M��*�����@a���cYX

qwen3-reranker-0.6b_float16_s512_static.aimodel/main.mlirb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26c4ef6f91695f161d6589e44da20b8d172a1ceaf2f6a1b34061fdd2e6635958
+size 1192490779

qwen3-reranker-0.6b_float16_s512_static.aimodel/metadata.json

@@ -0,0 +1,7 @@
+{
+  "description" : "Qwen3-Reranker-0.6B cross-encoder reranker (Qwen3-0.6B backbone; yes\/no logit score). Output probs[1] = P(yes) = relevance. Source: https:\/\/huggingface.co\/Qwen\/Qwen3-Reranker-0.6B",
+  "assetVersion" : "2.0",
+  "creationDate" : "20260614T052234Z",
+  "license" : "Apache-2.0",
+  "author" : "Alibaba Qwen"
+}

reference.json

@@ -0,0 +1,75 @@
+{
+  "model": "Qwen/Qwen3-Reranker-0.6B",
+  "seq_len": 512,
+  "dtype": "float16",
+  "yes_id": 9693,
+  "no_id": 2152,
+  "pad_token_id": 151643,
+  "padding_side": "right",
+  "prefix": "<|im_start|>system\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be \"yes\" or \"no\".<|im_end|>\n<|im_start|>user\n",
+  "suffix": "<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n",
+  "default_instruction": "Given a web search query, retrieve relevant passages that answer the query",
+  "output": "probs [1,2] = softmax([no, yes]); relevance = probs[1] = P(yes)",
+  "pairs": {
+    "rel_capital": {
+      "relevant": true,
+      "query": "What is the capital of Japan?",
+      "doc": "Tokyo is the capital and largest city of Japan."
+    },
+    "rel_beesting": {
+      "relevant": true,
+      "query": "How do I treat a bee sting?",
+      "doc": "Remove the stinger, wash with soap and water, then apply a cold pack to reduce swelling."
+    },
+    "rel_fuji_ja": {
+      "relevant": true,
+      "query": "富士山の高さはどのくらいですか？",
+      "doc": "富士山は標高3,776メートルで、日本で最も高い山です。"
+    },
+    "irr_capital": {
+      "relevant": false,
+      "query": "What is the capital of Japan?",
+      "doc": "Python is the most widely used programming language for machine learning."
+    },
+    "irr_beesting": {
+      "relevant": false,
+      "query": "How do I treat a bee sting?",
+      "doc": "Tokyo is the capital and largest city of Japan."
+    },
+    "irr_fuji_ja": {
+      "relevant": false,
+      "query": "富士山の高さはどのくらいですか？",
+      "doc": "The recipe calls for two eggs and a cup of flour."
+    }
+  },
+  "scores": {
+    "rel_capital": 0.9937748908996582,
+    "rel_beesting": 0.9814825654029846,
+    "rel_fuji_ja": 0.9997767806053162,
+    "irr_capital": 8.651458301756065e-06,
+    "irr_beesting": 3.616727553890087e-05,
+    "irr_fuji_ja": 5.813539701193804e-06
+  },
+  "official_scores": {
+    "rel_capital": 0.9937747716903687,
+    "rel_beesting": 0.9814824461936951,
+    "rel_fuji_ja": 0.9997767806053162,
+    "irr_capital": 8.651466487208381e-06,
+    "irr_beesting": 3.616731191868894e-05,
+    "irr_fuji_ja": 5.813562438561348e-06
+  },
+  "rank_groups": {
+    "capital": [
+      "rel_capital",
+      "irr_capital"
+    ],
+    "beesting": [
+      "rel_beesting",
+      "irr_beesting"
+    ],
+    "fuji_ja": [
+      "rel_fuji_ja",
+      "irr_fuji_ja"
+    ]
+  }
+}

tokenizer/chat_template.jinja

@@ -0,0 +1,15 @@
+{%- set instruction = messages | selectattr("role", "eq", "system") | map(attribute="content") | first | default("Given a web search query, retrieve relevant passages that answer the query") -%}
+{%- set query_text = messages | selectattr("role", "eq", "query") | map(attribute="content") | first -%}
+{%- set document_text = messages | selectattr("role", "eq", "document") | map(attribute="content") | first -%}
+<|im_start|>system
+Judge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be "yes" or "no".<|im_end|>
+<|im_start|>user
+<Instruct>: {{ instruction }}
+<Query>: {{ query_text }}
+<Document>: {{ document_text }}<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+

tokenizer/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650

tokenizer/tokenizer_config.json

@@ -0,0 +1,15 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}