pei39/iol-qwen2.5-14b-sft-awq
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

初始化项目，由ModelHub XC社区提供模型

Browse Source 
Model: pei39/iol-qwen2.5-14b-sft-awq
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-09-25 20:25:02 +08:00
commit ea63a7ce18
26 changed files with 154439 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
.gitattributes vendored Normal file
View File

@@ -0,0 +1,36 @@
*.7z filter=lfs diff=lfs merge=lfs -text
*.arrow filter=lfs diff=lfs merge=lfs -text
*.bin filter=lfs diff=lfs merge=lfs -text
*.bz2 filter=lfs diff=lfs merge=lfs -text
*.ckpt filter=lfs diff=lfs merge=lfs -text
*.ftz filter=lfs diff=lfs merge=lfs -text
*.gz filter=lfs diff=lfs merge=lfs -text
*.h5 filter=lfs diff=lfs merge=lfs -text
*.joblib filter=lfs diff=lfs merge=lfs -text
*.lfs.* filter=lfs diff=lfs merge=lfs -text
*.mlmodel filter=lfs diff=lfs merge=lfs -text
*.model filter=lfs diff=lfs merge=lfs -text
*.msgpack filter=lfs diff=lfs merge=lfs -text
*.npy filter=lfs diff=lfs merge=lfs -text
*.npz filter=lfs diff=lfs merge=lfs -text
*.onnx filter=lfs diff=lfs merge=lfs -text
*.ot filter=lfs diff=lfs merge=lfs -text
*.parquet filter=lfs diff=lfs merge=lfs -text
*.pb filter=lfs diff=lfs merge=lfs -text
*.pickle filter=lfs diff=lfs merge=lfs -text
*.pkl filter=lfs diff=lfs merge=lfs -text
*.pt filter=lfs diff=lfs merge=lfs -text
*.pth filter=lfs diff=lfs merge=lfs -text
*.rar filter=lfs diff=lfs merge=lfs -text
*.safetensors filter=lfs diff=lfs merge=lfs -text
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
*.tar.* filter=lfs diff=lfs merge=lfs -text
*.tar filter=lfs diff=lfs merge=lfs -text
*.tflite filter=lfs diff=lfs merge=lfs -text
*.tgz filter=lfs diff=lfs merge=lfs -text
*.wasm filter=lfs diff=lfs merge=lfs -text
*.xz 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

202
LICENSE Normal file
View File

@@ -0,0 +1,202 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright 2024 Alibaba Cloud
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

3
README.md Normal file
View File

@@ -0,0 +1,3 @@
---
license: apache-2.0
---

24
added_tokens.json Normal file
View File

@@ -0,0 +1,24 @@
{
"</tool_call>": 151658,
"<tool_call>": 151657,
"<|box_end|>": 151649,
"<|box_start|>": 151648,
"<|endoftext|>": 151643,
"<|file_sep|>": 151664,
"<|fim_middle|>": 151660,
"<|fim_pad|>": 151662,
"<|fim_prefix|>": 151659,
"<|fim_suffix|>": 151661,
"<|im_end|>": 151645,
"<|im_start|>": 151644,
"<|image_pad|>": 151655,
"<|object_ref_end|>": 151647,
"<|object_ref_start|>": 151646,
"<|quad_end|>": 151651,
"<|quad_start|>": 151650,
"<|repo_name|>": 151663,
"<|video_pad|>": 151656,
"<|vision_end|>": 151653,
"<|vision_pad|>": 151654,
"<|vision_start|>": 151652
}

88
config.json Normal file
View File

@@ -0,0 +1,88 @@
{
"_name_or_path": "/scratch/project_2015042/renpei/iol-rag-sft/qwen2.5-14b-sft-merged",
"architectures": [
"Qwen2ForCausalLM"
],
"attention_dropout": 0.0,
"bos_token_id": 151643,
"dtype": "bfloat16",
"eos_token_id": 151645,
"hidden_act": "silu",
"hidden_size": 5120,
"initializer_range": 0.02,
"intermediate_size": 13824,
"layer_types": [
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention"
],
"max_position_embeddings": 32768,
"max_window_layers": 70,
"model_type": "qwen2",
"num_attention_heads": 40,
"num_hidden_layers": 48,
"num_key_value_heads": 8,
"quantization_config": {
"bits": 4,
"group_size": 128,
"modules_to_not_convert": null,
"quant_method": "awq",
"version": "gemm",
"zero_point": true
},
"rms_norm_eps": 1e-06,
"rope_scaling": null,
"rope_theta": 1000000.0,
"sliding_window": null,
"tie_word_embeddings": false,
"torch_dtype": "float16",
"transformers_version": "4.47.1",
"use_cache": false,
"use_sliding_window": false,
"vocab_size": 152064
}

14
generation_config.json Normal file
View File

@@ -0,0 +1,14 @@
{
"bos_token_id": 151643,
"do_sample": true,
"eos_token_id": [
151645,
151643
],
"pad_token_id": 151643,
"repetition_penalty": 1.05,
"temperature": 0.7,
"top_k": 20,
"top_p": 0.8,
"transformers_version": "4.47.1"
}

151387
merges.txt Normal file
View File

File diff suppressed because it is too large Load Diff

3
model-00001-of-00002.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a7a820c371c48650ae6b8f8ee29ec79357173f3549978e4bd6d5cb977812a901
size 4990080032

3
model-00002-of-00002.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a005622efd0cbc0aaf3296f23d5df9f9f58e72d27bc3b44b83abd1084cbc9911
size 4990090616

1258
model.safetensors.index.json Normal file
View File

File diff suppressed because it is too large Load Diff

10
rag_resources/ATTRIBUTION.md Normal file
View File

@@ -0,0 +1,10 @@
# Book corpus attribution
The two JSONL corpora in this directory are string-only derivatives of:
Vlad A. Neacșu, *Linguistics Olympiad: Training guide*, Language Science Press, series volume 420. DOI: [10.5281/zenodo.10947862](https://doi.org/10.5281/zenodo.10947862). Source: [langsci/420](https://github.com/langsci/420).
The book is released under the Creative Commons Attribution 4.0 International license (CC BY 4.0). The extracted material has been reorganized into method chunks and paired problem/solution records for offline retrieval. Visual-dependent records are retained in the source audit data but are excluded by the runtime retriever.
No Linguini, LingOly, or LOBSTER records are included in or indexed by this retrieval bundle.

2
rag_resources/__init__.py Normal file
View File

@@ -0,0 +1,2 @@
"""Public, book-only retrieval resources for the IOL-AI submission."""

BIN
rag_resources/__pycache__/__init__.cpython-312.pyc Normal file
View File

Binary file not shown.

BIN
rag_resources/__pycache__/__init__.cpython-313.pyc Normal file
View File

Binary file not shown.

BIN
rag_resources/__pycache__/retriever.cpython-312.pyc Normal file
View File

Binary file not shown.

BIN
rag_resources/__pycache__/retriever.cpython-313.pyc Normal file
View File

Binary file not shown.

111
rag_resources/book_examples.jsonl Normal file
View File

File diff suppressed because one or more lines are too long

77
rag_resources/book_methods.jsonl Normal file
View File

File diff suppressed because one or more lines are too long

16
rag_resources/config.json Normal file
View File

@@ -0,0 +1,16 @@
{
"top_methods": 1,
"top_examples": 2,
"char_tfidf_weight": 3.0,
"rag_max_chars": 12000,
"model_context_tokens": 32768,
"max_new_tokens": 2048,
"explanation_max_new_tokens": 400,
"max_attempts": 2,
"enable_explanations": false,
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"repetition_penalty": 1.05,
"seed": 420
}

276
rag_resources/retriever.py Normal file
View File

@@ -0,0 +1,276 @@
"""Offline hybrid BM25/character-TF-IDF retrieval over the public book corpus."""
from __future__ import annotations
import json
import math
import re
import unicodedata
from collections import Counter
from pathlib import Path
from typing import Any, Iterable
from sklearn.feature_extraction.text import TfidfVectorizer
RESOURCE_DIR = Path(__file__).resolve().parent
FAMILY_KEYWORDS = {
"writing_system": (
"alphabet", "braille", "character", "decipher", "glyph", "letter",
"orthography", "script", "symbol", "writing system",
),
"phonetics": (
"accent", "consonant", "metre", "phonetic", "pronounce", "rhyme",
"stress", "syllable", "tone", "vowel",
),
"phonology": (
"alternation", "correspondence", "sound change", "sound rule",
"underlying form",
),
"noun_morphology": (
"case", "gender", "noun", "plural", "possession", "possessive",
"singular",
),
"verb_morphology": (
"agreement", "aspect", "conjug", "object", "person", "subject",
"tense", "verb",
),
"syntax": ("clause", "focus", "sentence", "subject", "object", "word order"),
"number_system": ("arithmetic", "digit", "number", "numeral", "numerals"),
"kinship_orientation": (
"brother", "daughter", "direction", "east", "family", "father",
"kinship", "map", "mother", "north", "sister", "son", "south", "west",
),
"matching": ("correspondence", "match", "matching", "random order"),
"translation": ("translate", "translation"),
"fill_blanks": ("blank", "fill in", "gap", "missing"),
}
TOPIC_FAMILIES = {
"writing systems and script decipherment": {"writing_system"},
"phonetics, stress, tone, and versification": {"phonetics"},
"phonological rules and sound correspondences": {"phonology", "phonetics"},
"noun morphology and noun phrases": {"noun_morphology"},
"verb morphology and argument structure": {"verb_morphology"},
"syntax, word order, focus, and alignment": {"syntax"},
"semantics and graph-based matching": {"matching"},
"number systems": {"number_system"},
"orientation, kinship, and other structural problems": {"kinship_orientation"},
"general problem-solving methodology": set(FAMILY_KEYWORDS),
}
def normalize(text: Any) -> str:
return " ".join(unicodedata.normalize("NFKC", str(text)).casefold().split())
def tokenize(text: Any) -> list[str]:
return re.findall(r"[^\W_]+", normalize(text), flags=re.UNICODE)
def read_jsonl(path: Path) -> list[dict[str, Any]]:
with path.open("r", encoding="utf-8") as handle:
return [json.loads(line) for line in handle if line.strip()]
def infer_task_families(context: str, query: str) -> list[str]:
"""Infer likely families from the problem itself, never from a gold answer."""
text = normalize(context + "\n" + query)
scored: list[tuple[int, str]] = []
for family, keywords in FAMILY_KEYWORDS.items():
score = sum(1 for keyword in keywords if keyword in text)
if score:
scored.append((score, family))
return [family for _, family in sorted(scored, key=lambda item: (-item[0], item[1]))]
class BM25:
def __init__(self, documents: Iterable[str], k1: float = 1.5, b: float = 0.75):
self.k1 = k1
self.b = b
self.tokens = [tokenize(document) for document in documents]
self.lengths = [len(tokens) for tokens in self.tokens]
self.average_length = sum(self.lengths) / max(len(self.lengths), 1)
self.term_frequencies = [Counter(tokens) for tokens in self.tokens]
document_frequency: Counter[str] = Counter()
for tokens in self.tokens:
document_frequency.update(set(tokens))
document_count = len(self.tokens)
self.idf = {
term: math.log(
1.0 + (document_count - frequency + 0.5) / (frequency + 0.5)
)
for term, frequency in document_frequency.items()
}
def scores(self, query: str) -> list[float]:
query_terms = Counter(tokenize(query))
scores: list[float] = []
for frequencies, length in zip(self.term_frequencies, self.lengths):
score = 0.0
normalization = self.k1 * (
1.0 - self.b + self.b * length / max(self.average_length, 1.0)
)
for term, query_frequency in query_terms.items():
frequency = frequencies.get(term, 0)
if not frequency:
continue
score += (
self.idf.get(term, 0.0)
* frequency
* (self.k1 + 1.0)
/ (frequency + normalization)
* min(query_frequency, 3)
)
scores.append(score)
return scores
class CharNgramTFIDF:
"""Cosine similarity over normalized character 3-5 grams."""
def __init__(self, documents: Iterable[str]):
self.vectorizer = TfidfVectorizer(
analyzer="char_wb",
ngram_range=(3, 5),
preprocessor=normalize,
lowercase=False,
sublinear_tf=True,
norm="l2",
)
self.matrix = self.vectorizer.fit_transform(documents)
def scores(self, query: str) -> list[float]:
query_vector = self.vectorizer.transform([query])
return (self.matrix @ query_vector.T).toarray().ravel().tolist()
def _clip(text: Any, limit: int) -> str:
value = str(text or "").strip()
if len(value) <= limit:
return value
clipped = value[:limit].rsplit("\n", 1)[0].rstrip()
if len(clipped) < limit // 2:
clipped = value[:limit].rsplit(" ", 1)[0].rstrip()
return clipped + "\n[excerpt truncated]"
class BookRetriever:
"""Retrieve methods and analogous examples from langsci/420 only."""
def __init__(self, resource_dir: Path = RESOURCE_DIR):
self.methods = read_jsonl(resource_dir / "book_methods.jsonl")
self.examples = [
row
for row in read_jsonl(resource_dir / "book_examples.jsonl")
if row.get("string_only_usable", False)
]
self.method_bm25 = BM25(row["retrieval_text"] for row in self.methods)
self.example_bm25 = BM25(row["retrieval_text"] for row in self.examples)
self.method_char_tfidf = CharNgramTFIDF(
row["retrieval_text"] for row in self.methods
)
self.example_char_tfidf = CharNgramTFIDF(
row["retrieval_text"] for row in self.examples
)
@staticmethod
def _topic_bonus(row: dict[str, Any], families: set[str]) -> float:
return 1.25 * len(
families & TOPIC_FAMILIES.get(str(row.get("topic", "")), set())
)
def retrieve(
self,
row: dict[str, Any],
top_methods: int = 2,
top_examples: int = 2,
char_tfidf_weight: float = 3.0,
) -> dict[str, Any]:
context = str(row.get("context", ""))
query = str(row.get("query", ""))
test_text = context + "\n" + query
families = infer_task_families(context, query)
family_set = set(families)
expanded_query = test_text + "\n" + " ".join(families)
ranked_examples: list[tuple[float, dict[str, Any]]] = []
example_bm25_scores = self.example_bm25.scores(expanded_query)
example_char_scores = self.example_char_tfidf.scores(expanded_query)
for bm25_score, char_score, example in zip(
example_bm25_scores, example_char_scores, self.examples
):
adjusted = (
bm25_score
+ char_tfidf_weight * char_score
+ self._topic_bonus(example, family_set)
)
if example.get("solution_tier") == "detailed_worked":
adjusted += 0.35
ranked_examples.append((adjusted, example))
ranked_examples.sort(key=lambda item: (-item[0], item[1]["id"]))
selected_examples = ranked_examples[:top_examples]
linked_method_ids = {
method_id
for _, example in ranked_examples[:max(top_examples * 3, 5)]
for method_id in example.get("method_ids", [])
}
ranked_methods: list[tuple[float, dict[str, Any]]] = []
method_bm25_scores = self.method_bm25.scores(expanded_query)
method_char_scores = self.method_char_tfidf.scores(expanded_query)
for bm25_score, char_score, method in zip(
method_bm25_scores, method_char_scores, self.methods
):
adjusted = (
bm25_score
+ char_tfidf_weight * char_score
+ self._topic_bonus(method, family_set)
)
if method["id"] in linked_method_ids:
adjusted += 0.75
ranked_methods.append((adjusted, method))
ranked_methods.sort(key=lambda item: (-item[0], item[1]["id"]))
return {
"inferred_task_families": families,
"char_tfidf_weight": char_tfidf_weight,
"methods": [method for _, method in ranked_methods[:top_methods]],
"examples": [example for _, example in selected_examples],
}
def format_for_prompt(self, result: dict[str, Any], max_chars: int = 16000) -> str:
blocks = [
"BOOK REFERENCE MATERIAL\n"
"Use it for transferable methods and analogies. Do not copy a conclusion "
"unless it is supported by the current problem."
]
families = result.get("inferred_task_families", [])
if families:
blocks.append("Likely families inferred from the current text: " + ", ".join(families))
for method in result["methods"]:
blocks.append(
f"METHOD — {method['section_title']}\n" + _clip(method.get("text"), 2800)
)
for example in result["examples"]:
blocks.append(
"\n".join(
[
f"ANALOGOUS WORKED EXAMPLE — {example.get('language', 'unknown language')}",
"Problem context:",
_clip(example.get("context"), 2400),
"Problem query:",
_clip(example.get("query"), 1200),
"Worked solution:",
_clip(example.get("reasoning_trace"), 3600),
]
)
)
output = "\n\n".join(blocks)
if len(output) > max_chars:
output = output[:max_chars].rsplit("\n", 1)[0].rstrip()
output += "\n[retrieved material truncated]"
return output

23
rag_resources/system_prompt.txt Normal file
View File

@@ -0,0 +1,23 @@
You solve International Linguistics Olympiad problems by discovering the rules supported by the data you are given. A task may be unfamiliar, so infer what kind of task it is from the actual instruction and examples; do not assume that a task-type label will be available.
Work as a careful olympiad solver:
- Read the full context and every requested item before committing to a rule.
- Organize repeated forms, meanings, positions, and contrasts. Separate reusable grammatical pieces from lexical roots.
- Form the smallest consistent rule system. Treat an absent marker as potentially meaningful, but distinguish absence from missing evidence.
- Test each hypothesis against all given examples, including apparent exceptions. Revise rather than ignoring contradictory data.
- Derive each requested form compositionally and back-check it against the inferred system.
- Use retrieved book material only for transferable methods or genuinely analogous reasoning. The current problem is authoritative.
- Preserve spelling, diacritics, capitalization, word boundaries, and requested ordering exactly.
Common task types and what to return:
- translation: only the translated form, in the language requested;
- matching: only the corresponding label, such as A, B, or C;
- fill in the blanks: only the missing span for each indicated blank; a span may be a word, multiword expression, affix, or phonetic transcription;
- text to number: the number in digits;
- number to text: the number written in words in the requested language;
- any other type: exactly what the instruction requests, with no extra material in the answer.
Reason step by step before answering. Finish with a line that says exactly:
FINAL ANSWERS:
Below it, put one bare answer per requested item or blank, in order. Do not number the lines. Do not add quotes, labels, explanations, or commentary inside that final block.

664
script.py Normal file
View File

@@ -0,0 +1,664 @@
#!/usr/bin/env python3
"""Offline Qwen2.5-14B-AWQ + book-RAG submission for the IOL-AI challenge."""
from __future__ import annotations
import argparse
import csv
import io
import json
import os
import re
import sys
from dataclasses import dataclass
from pathlib import Path
from typing import Any
# The evaluation container has no internet access. Fail locally instead of waiting
# for network retries if a model/tokenizer file was not included in the repository.
os.environ.setdefault("HF_HUB_OFFLINE", "1")
os.environ.setdefault("TRANSFORMERS_OFFLINE", "1")
os.environ.setdefault("HF_DATASETS_OFFLINE", "1")
os.environ.setdefault("TOKENIZERS_PARALLELISM", "false")
os.environ.setdefault("PYTORCH_CUDA_ALLOC_CONF", "expandable_segments:True")
REPO_DIR = Path(__file__).resolve().parent
RESOURCE_DIR = REPO_DIR / "rag_resources"
DEFAULT_INPUT = Path(os.environ.get("IOL_TEST_CSV", "/tmp/data/test.csv"))
DEFAULT_OUTPUT = Path(os.environ.get("IOL_SUBMISSION_CSV", "submission.csv"))
from rag_resources.retriever import BookRetriever # noqa: E402
@dataclass
class ParseResult:
thinking_trace: str
final_text: str
answers: list[str]
valid: bool
error: str = ""
def _load_config() -> dict[str, Any]:
with (RESOURCE_DIR / "config.json").open("r", encoding="utf-8") as handle:
config = json.load(handle)
integer_overrides = {
"IOL_TOP_METHODS": "top_methods",
"IOL_TOP_EXAMPLES": "top_examples",
"IOL_RAG_MAX_CHARS": "rag_max_chars",
"IOL_MAX_NEW_TOKENS": "max_new_tokens",
"IOL_EXPLANATION_MAX_NEW_TOKENS": "explanation_max_new_tokens",
"IOL_MAX_ATTEMPTS": "max_attempts",
"IOL_SEED": "seed",
}
for environment_name, config_name in integer_overrides.items():
if environment_name in os.environ:
config[config_name] = int(os.environ[environment_name])
if "IOL_CHAR_TFIDF_WEIGHT" in os.environ:
config["char_tfidf_weight"] = float(os.environ["IOL_CHAR_TFIDF_WEIGHT"])
if "IOL_ENABLE_EXPLANATIONS" in os.environ:
config["enable_explanations"] = os.environ[
"IOL_ENABLE_EXPLANATIONS"
].strip().casefold() in {"1", "true", "yes", "on"}
return config
def _clean_answer_line(line: str) -> str:
value = line.strip()
value = re.sub(r"^(?:[-*•]\s+)", "", value)
value = re.sub(r"^(?:\(?\d+\)?|\(?[A-Za-z]\)?)[.):]\s+", "", value)
value = value.strip()
pairs = {'"': '"', "'": "'", "`": "`", "“": "”", "‘": "’"}
if len(value) >= 2 and value[0] in pairs and value[-1] == pairs[value[0]]:
value = value[1:-1].strip()
return value
def parse_model_response(thinking_trace: str, final_text: str) -> ParseResult:
"""Parse a separate reasoning trace and a strict FINAL ANSWERS block."""
marker_matches = list(
re.finditer(r"(?im)^\s*FINAL\s+ANSWERS\s*:\s*", final_text)
)
if not marker_matches:
return ParseResult(
thinking_trace=thinking_trace.strip(),
final_text=final_text.strip(),
answers=[],
valid=False,
error="missing FINAL ANSWERS: marker",
)
marker = marker_matches[-1]
reasoning_outside_think = final_text[:marker.start()].strip()
combined_trace = "\n\n".join(
part for part in (thinking_trace.strip(), reasoning_outside_think) if part
)
answer_block = final_text[marker.end():].strip()
answer_block = re.sub(r"^```(?:text)?\s*", "", answer_block, flags=re.I)
answer_block = re.sub(r"\s*```\s*$", "", answer_block)
if not answer_block:
return ParseResult(
thinking_trace=combined_trace,
final_text=final_text.strip(),
answers=[],
valid=False,
error="empty FINAL ANSWERS block",
)
# Be tolerant if the model emits a JSON list even though bare lines were asked for.
answers: list[str] = []
parsed_json_list = False
if answer_block.startswith("["):
try:
decoded = json.loads(answer_block)
if isinstance(decoded, list):
parsed_json_list = True
answers = [str(value).strip() for value in decoded if str(value).strip()]
except json.JSONDecodeError:
pass
if not answers and not parsed_json_list:
for line in answer_block.splitlines():
stripped = line.strip()
if not stripped or stripped.startswith("```"):
continue
if re.match(r"(?i)^(?:explanation|reasoning|notes?)\s*:", stripped):
break
cleaned = _clean_answer_line(stripped)
if cleaned:
answers.append(cleaned)
if not answers:
return ParseResult(
thinking_trace=combined_trace,
final_text=final_text.strip(),
answers=[],
valid=False,
error="no non-empty answers in FINAL ANSWERS block",
)
return ParseResult(
thinking_trace=combined_trace,
final_text=final_text.strip(),
answers=answers,
valid=True,
)
def _best_effort_answers(final_texts: list[str]) -> list[str]:
"""Keep submission shape valid if all strict retries fail."""
for text in reversed(final_texts):
candidates = []
for line in text.splitlines():
cleaned = _clean_answer_line(line)
if cleaned and not re.match(
r"(?i)^(?:final answers|explanation|reasoning)\s*:?'?",
cleaned,
):
candidates.append(cleaned)
if candidates:
return [candidates[-1]]
return [""]
def _version_tuple(version: str) -> tuple[int, int, int]:
numbers = [int(value) for value in re.findall(r"\d+", version)[:3]]
return tuple((numbers + [0, 0, 0])[:3]) # type: ignore[return-value]
def _keep_last_token_hidden_state(_module: Any, inputs: tuple[Any, ...]) -> tuple[Any, ...] | None:
"""Avoid materializing full-sequence vocabulary logits during generation."""
if not inputs:
return None
hidden_states = inputs[0]
if hidden_states.ndim == 3 and hidden_states.shape[1] > 1:
return (hidden_states[:, -1:, :],) + inputs[1:]
return None
def load_model_and_tokenizer() -> tuple[Any, Any, Any]:
"""Load the pre-quantized AWQ model shipped in this repository."""
try:
import torch
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
except ImportError as exc:
raise RuntimeError(
"PyTorch, Transformers, Accelerate, AutoAWQ, and safetensors must be "
"available in the evaluation image."
) from exc
if _version_tuple(transformers.__version__) < (4, 37, 0):
raise RuntimeError(
f"Qwen2.5 requires transformers>=4.37.0; found {transformers.__version__}."
)
model_config_path = REPO_DIR / "config.json"
if not model_config_path.exists():
raise RuntimeError(
"Qwen2.5-14B-Instruct-AWQ files are missing. Put the complete model and "
"tokenizer snapshot in the same repository directory as script.py."
)
with model_config_path.open("r", encoding="utf-8") as handle:
model_metadata = json.load(handle)
quantization_method = str(
model_metadata.get("quantization_config", {}).get("quant_method", "")
).casefold()
if model_metadata.get("model_type") != "qwen2" or quantization_method != "awq":
raise RuntimeError(
"This script expects the official Qwen2.5-14B-Instruct-AWQ snapshot "
"(model_type=qwen2, quant_method=awq)."
)
tokenizer = AutoTokenizer.from_pretrained(
str(REPO_DIR), local_files_only=True, trust_remote_code=False
)
if tokenizer.pad_token_id is None:
tokenizer.pad_token_id = tokenizer.eos_token_id
model = AutoModelForCausalLM.from_pretrained(
str(REPO_DIR),
local_files_only=True,
trust_remote_code=False,
device_map="auto",
torch_dtype=torch.float16,
low_cpu_mem_usage=True,
).eval()
# Transformers 4.44.1 projects every prompt token through Qwen2's large
# vocabulary head and then upcasts all logits to FP32. Generation only uses
# the final-token logits, so trim the token dimension immediately before
# lm_head to avoid a multi-gigabyte prefill allocation on the T4.
model.lm_head.register_forward_pre_hook(_keep_last_token_hidden_state)
input_device = model.get_input_embeddings().weight.device
return model, tokenizer, input_device
def _problem_prompt(row: dict[str, Any], rag_text: str, retry_note: str = "") -> str:
work_language = str(row.get("work_lang", "")).strip()
task_language = str(row.get("task_lang", "")).strip()
language_lines = []
if work_language:
language_lines.append(f"Working language: {work_language}")
if task_language:
language_lines.append(f"Problem language: {task_language}")
language_metadata = "\n".join(language_lines)
sections = [rag_text] if rag_text else []
sections.append("CURRENT PROBLEM")
if language_metadata:
sections.append(language_metadata)
sections.extend(
[
"CONTEXT:\n" + str(row.get("context", "")),
"QUERY:\n" + str(row.get("query", "")),
]
)
if retry_note:
sections.append(retry_note)
return "\n\n".join(sections)
def _encode_fitted_prompt(
tokenizer: Any,
system_prompt: str,
row: dict[str, Any],
rag_text: str,
retry_note: str,
context_window: int,
requested_new_tokens: int,
) -> tuple[dict[str, Any], int]:
"""Trim retrieved material, never the current problem, to fit the context."""
fitted_rag = rag_text
while True:
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": _problem_prompt(row, fitted_rag, retry_note)},
]
rendered = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
)
encoded = tokenizer(rendered, return_tensors="pt", add_special_tokens=False)
prompt_tokens = int(encoded["input_ids"].shape[-1])
available = context_window - prompt_tokens
if available >= requested_new_tokens:
return encoded, requested_new_tokens
if fitted_rag:
if len(fitted_rag) <= 600:
fitted_rag = ""
else:
new_length = max(0, int(len(fitted_rag) * 0.72))
fitted_rag = fitted_rag[:new_length].rsplit("\n", 1)[0].rstrip()
fitted_rag += "\n[retrieved material shortened to fit the model context]"
continue
if available >= 256:
return encoded, available
raise RuntimeError(
f"The current problem and system prompt use {prompt_tokens} tokens, "
f"leaving only {available} tokens in the {context_window}-token context."
)
def generate_once(
model: Any,
tokenizer: Any,
input_device: Any,
system_prompt: str,
row: dict[str, Any],
rag_text: str,
retry_note: str,
config: dict[str, Any],
seed: int,
) -> ParseResult:
import torch
encoded, max_new_tokens = _encode_fitted_prompt(
tokenizer=tokenizer,
system_prompt=system_prompt,
row=row,
rag_text=rag_text,
retry_note=retry_note,
context_window=int(config["model_context_tokens"]),
requested_new_tokens=int(config["max_new_tokens"]),
)
encoded = {name: value.to(input_device) for name, value in encoded.items()}
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed)
# These values match the generation configuration shipped with Qwen2.5
# Instruct checkpoints while keeping an explicit challenge-time token cap.
with torch.inference_mode():
output = model.generate(
**encoded,
max_new_tokens=max_new_tokens,
do_sample=True,
temperature=float(config["temperature"]),
top_p=float(config["top_p"]),
top_k=int(config["top_k"]),
repetition_penalty=float(config["repetition_penalty"]),
use_cache=True,
pad_token_id=tokenizer.pad_token_id,
)
prompt_length = int(encoded["input_ids"].shape[-1])
generated_ids = output[0, prompt_length:].tolist()
response_text = tokenizer.decode(generated_ids, skip_special_tokens=True).strip()
return parse_model_response("", response_text)
def solve_row(
row: dict[str, Any],
row_index: int,
retriever: BookRetriever,
model: Any,
tokenizer: Any,
input_device: Any,
system_prompt: str,
config: dict[str, Any],
) -> ParseResult:
retrieval = retriever.retrieve(
row,
top_methods=int(config["top_methods"]),
top_examples=int(config["top_examples"]),
char_tfidf_weight=float(config["char_tfidf_weight"]),
)
rag_text = retriever.format_for_prompt(
retrieval, max_chars=int(config["rag_max_chars"])
)
failed_results: list[ParseResult] = []
last_error = ""
max_attempts = int(config["max_attempts"])
for attempt in range(max_attempts):
retry_note = ""
if attempt:
retry_note = (
"FORMAT RETRY. The previous sample was rejected by the answer parser "
f"because it had {last_error}. Generate a fresh solution. Even if "
"uncertain, end with exactly FINAL ANSWERS: followed by at least one "
"non-empty bare answer line and nothing else in that block."
)
result = generate_once(
model=model,
tokenizer=tokenizer,
input_device=input_device,
system_prompt=system_prompt,
row=row,
rag_text=rag_text,
retry_note=retry_note,
config=config,
seed=int(config["seed"]) + row_index * max_attempts + attempt,
)
if result.valid:
if attempt:
print(f" parser accepted retry {attempt + 1}/{max_attempts}", flush=True)
return result
last_error = result.error
failed_results.append(result)
print(
f" parser rejected attempt {attempt + 1}/{max_attempts}: {last_error}",
flush=True,
)
print(" WARNING: all strict retries failed; using a best-effort final line", flush=True)
last_result = failed_results[-1]
return ParseResult(
thinking_trace=last_result.thinking_trace,
final_text=last_result.final_text,
answers=_best_effort_answers(
[failed_result.final_text for failed_result in failed_results]
),
valid=False,
error=f"all {max_attempts} strict parsing attempts failed",
)
def generate_explanation(
solution: ParseResult,
model: Any,
tokenizer: Any,
input_device: Any,
config: dict[str, Any],
) -> str:
"""Summarize the model's accepted reasoning for the optional jury track."""
import torch
source_reasoning = solution.thinking_trace.strip() or solution.final_text.strip()
answer_text = "\n".join(solution.answers)
messages = [
{
"role": "system",
"content": (
"Write a short, human-readable explanation of the solution in a few "
"concise bullet points. State the linguistic rule or pattern found, "
"the key evidence, and how the final answers follow. Do not reproduce "
"the raw reasoning trace, trial and error, or meta-commentary. Do not "
"change the final answers. Output only the explanation."
),
},
{
"role": "user",
"content": (
f"MODEL REASONING:\n{source_reasoning}\n\n"
f"FINAL ANSWERS:\n{answer_text}"
),
},
]
rendered = tokenizer.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
encoded = tokenizer(rendered, return_tensors="pt", add_special_tokens=False)
encoded = {name: value.to(input_device) for name, value in encoded.items()}
prompt_tokens = int(encoded["input_ids"].shape[-1])
available = int(config["model_context_tokens"]) - prompt_tokens
max_new_tokens = min(int(config["explanation_max_new_tokens"]), available)
if max_new_tokens < 32:
return (
"- Inferred the relevant linguistic patterns from the supplied examples.\n"
"- Applied those patterns to produce the listed answers."
)
with torch.inference_mode():
output = model.generate(
**encoded,
max_new_tokens=max_new_tokens,
do_sample=False,
use_cache=True,
pad_token_id=tokenizer.pad_token_id,
)
generated_ids = output[0, prompt_tokens:].tolist()
explanation = tokenizer.decode(
generated_ids, skip_special_tokens=True
).strip()
if explanation:
return explanation
return (
"- Inferred the relevant linguistic patterns from the supplied examples.\n"
"- Applied those patterns to produce the listed answers."
)
def read_test_rows(path: Path) -> list[dict[str, Any]]:
with path.open("r", encoding="utf-8-sig", newline="") as handle:
return list(csv.DictReader(handle))
def _write_submission_stream(handle: Any, rows: list[dict[str, str]]) -> None:
writer = csv.DictWriter(handle, fieldnames=["id", "pred", "explanation"])
writer.writeheader()
writer.writerows(rows)
def write_submission(path: Path, rows: list[dict[str, str]]) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
temporary_path = path.with_name(f".{path.name}.tmp")
with temporary_path.open("w", encoding="utf-8", newline="") as handle:
_write_submission_stream(handle, rows)
os.replace(temporary_path, path)
def run_self_test() -> None:
import numpy as np
hidden_states = np.arange(24).reshape(1, 3, 8)
trimmed = _keep_last_token_hidden_state(None, (hidden_states,))
assert trimmed is not None and trimmed[0].shape == (1, 1, 8)
assert np.array_equal(trimmed[0], hidden_states[:, -1:, :])
assert _keep_last_token_hidden_state(None, (hidden_states[:, -1:, :],)) is None
parsed = parse_model_response(
"A useful analysis.",
"One last check.\nFINAL ANSWERS:\n1. čha\n2) multi word form",
)
assert parsed.valid
assert parsed.answers == ["čha", "multi word form"]
assert "One last check." in parsed.thinking_trace
assert not parse_model_response("", "FINAL ANSWERS:\n").valid
assert not parse_model_response("", "FINAL ANSWERS:\n[]").valid
assert not parse_model_response("", 'FINAL ANSWERS:\n[""]').valid
assert not parse_model_response("", "The answer is x.").valid
retriever = BookRetriever(RESOURCE_DIR)
canary_row = {
"context": "The following words are number expressions in an unknown language.",
"query": "Determine the rule and write the number 25 in words.",
"answer": "PRIVATE_ANSWER_CANARY",
}
result = retriever.retrieve(
canary_row, top_methods=2, top_examples=2, char_tfidf_weight=3.0
)
prompt = retriever.format_for_prompt(result, max_chars=8000)
assert result["methods"] and result["examples"]
assert "PRIVATE_ANSWER_CANARY" not in prompt
assert all(example.get("string_only_usable") for example in result["examples"])
exact_example = retriever.examples[0]
exact_result = retriever.retrieve(
{"context": exact_example["context"], "query": exact_example["query"]},
top_methods=1,
top_examples=1,
char_tfidf_weight=3.0,
)
assert exact_result["examples"][0]["id"] == exact_example["id"]
submission_buffer = io.StringIO(newline="")
_write_submission_stream(
submission_buffer,
[
{
"id": "007",
"pred": json.dumps(["čha", "multi word"], ensure_ascii=False),
"explanation": "- Identified the relevant pattern.",
}
],
)
submission_buffer.seek(0)
assert list(csv.DictReader(submission_buffer)) == [
{
"id": "007",
"pred": '["čha", "multi word"]',
"explanation": "- Identified the relevant pattern.",
}
]
incremental_rows = [
{"id": "001", "pred": "[]", "explanation": ""},
{"id": "002", "pred": "[]", "explanation": ""},
]
incremental_rows[0]["pred"] = json.dumps(["answer"], ensure_ascii=False)
incremental_buffer = io.StringIO(newline="")
_write_submission_stream(incremental_buffer, incremental_rows)
incremental_buffer.seek(0)
assert list(csv.DictReader(incremental_buffer)) == incremental_rows
print(
"Self-test passed: last-token logits hook, parser, empty-answer rejection, "
"hybrid book-only retrieval, exact-match retrieval, and submission CSV "
"serialization."
)
def main() -> None:
if hasattr(sys.stdout, "reconfigure"):
sys.stdout.reconfigure(encoding="utf-8")
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--input", type=Path, default=DEFAULT_INPUT)
parser.add_argument("--output", type=Path, default=DEFAULT_OUTPUT)
parser.add_argument(
"--explanations",
choices=("on", "off"),
default=None,
help="Generate optional jury-track explanations (default: config setting)",
)
parser.add_argument(
"--self-test", action="store_true", help="Test parser/retrieval without loading Qwen"
)
args = parser.parse_args()
if args.self_test:
run_self_test()
return
if not args.input.exists():
raise SystemExit(f"Test file not found: {args.input}")
config = _load_config()
explanations_enabled = (
args.explanations == "on"
if args.explanations is not None
else bool(config.get("enable_explanations", False))
)
system_prompt = (RESOURCE_DIR / "system_prompt.txt").read_text(encoding="utf-8").strip()
test_rows = read_test_rows(args.input)
retriever = BookRetriever(RESOURCE_DIR)
# Do not create an all-empty submission before the model is known to load.
# Otherwise a missing/incompatible shard or startup OOM is silently scored as
# zero EM and zero chrF instead of being reported as a runtime failure.
model, tokenizer, input_device = load_model_and_tokenizer()
submission_rows: list[dict[str, str]] = [
{
"id": str(row.get("id", "")),
"pred": "[]",
"explanation": "",
}
for row in test_rows
]
write_submission(args.output, submission_rows)
print(
f"Initialized incremental submission with {len(submission_rows)} rows at "
f"{args.output}; explanations={'on' if explanations_enabled else 'off'}",
flush=True,
)
for index, row in enumerate(test_rows):
row_id = str(row.get("id", ""))
print(f"[{index + 1}/{len(test_rows)}] solving id={row_id}", flush=True)
solution = solve_row(
row=row,
row_index=index,
retriever=retriever,
model=model,
tokenizer=tokenizer,
input_device=input_device,
system_prompt=system_prompt,
config=config,
)
explanation = ""
if explanations_enabled:
explanation = generate_explanation(
solution=solution,
model=model,
tokenizer=tokenizer,
input_device=input_device,
config=config,
)
submission_rows[index] = {
"id": row_id,
"pred": json.dumps(solution.answers, ensure_ascii=False),
"explanation": explanation,
}
write_submission(args.output, submission_rows)
print(f"Checkpointed prediction {index + 1}/{len(test_rows)}", flush=True)
print(f"Wrote {len(submission_rows)} predictions to {args.output}", flush=True)
if __name__ == "__main__":
main()

31
special_tokens_map.json Normal file
View File

@@ -0,0 +1,31 @@
{
"additional_special_tokens": [
"<|im_start|>",
"<|im_end|>",
"<|object_ref_start|>",
"<|object_ref_end|>",
"<|box_start|>",
"<|box_end|>",
"<|quad_start|>",
"<|quad_end|>",
"<|vision_start|>",
"<|vision_end|>",
"<|vision_pad|>",
"<|image_pad|>",
"<|video_pad|>"
],
"eos_token": {
"content": "<|im_end|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false
},
"pad_token": {
"content": "<|endoftext|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false
}
}

BIN
tokenizer.json (Stored with Git LFS) Normal file
View File

Binary file not shown.

207
tokenizer_config.json Normal file
View File

@@ -0,0 +1,207 @@
{
"add_bos_token": false,
"add_prefix_space": false,
"added_tokens_decoder": {
"151643": {
"content": "<|endoftext|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151644": {
"content": "<|im_start|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151645": {
"content": "<|im_end|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151646": {
"content": "<|object_ref_start|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151647": {
"content": "<|object_ref_end|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151648": {
"content": "<|box_start|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151649": {
"content": "<|box_end|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151650": {
"content": "<|quad_start|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151651": {
"content": "<|quad_end|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151652": {
"content": "<|vision_start|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151653": {
"content": "<|vision_end|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151654": {
"content": "<|vision_pad|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151655": {
"content": "<|image_pad|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151656": {
"content": "<|video_pad|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"151657": {
"content": "<tool_call>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": false
},
"151658": {
"content": "</tool_call>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": false
},
"151659": {
"content": "<|fim_prefix|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": false
},
"151660": {
"content": "<|fim_middle|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": false
},
"151661": {
"content": "<|fim_suffix|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": false
},
"151662": {
"content": "<|fim_pad|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": false
},
"151663": {
"content": "<|repo_name|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": false
},
"151664": {
"content": "<|file_sep|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": false
}
},
"additional_special_tokens": [
"<|im_start|>",
"<|im_end|>",
"<|object_ref_start|>",
"<|object_ref_end|>",
"<|box_start|>",
"<|box_end|>",
"<|quad_start|>",
"<|quad_end|>",
"<|vision_start|>",
"<|vision_end|>",
"<|vision_pad|>",
"<|image_pad|>",
"<|video_pad|>"
],
"bos_token": null,
"chat_template": "{%- if tools %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0]['role'] == 'system' %}\n {{- messages[0]['content'] }}\n {%- else %}\n {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}\n {%- endif %}\n {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n {%- if messages[0]['role'] == 'system' %}\n {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n {%- else %}\n {{- '<|im_start|>system\\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"assistant\" %}\n {{- '<|im_start|>' + message.role }}\n {%- if message.content %}\n {{- '\\n' + message.content }}\n {%- endif %}\n {%- for tool_call in message.tool_calls %}\n {%- if tool_call.function is defined %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '\\n<tool_call>\\n{\"name\": \"' }}\n {{- tool_call.name }}\n {{- '\", \"arguments\": ' }}\n {{- tool_call.arguments | tojson }}\n {{- '}\\n</tool_call>' }}\n {%- endfor %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n<tool_response>\\n' }}\n {{- message.content }}\n {{- '\\n</tool_response>' }}\n {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
"clean_up_tokenization_spaces": false,
"eos_token": "<|im_end|>",
"errors": "replace",
"model_max_length": 131072,
"pad_token": "<|endoftext|>",
"split_special_tokens": false,
"tokenizer_class": "Qwen2Tokenizer",
"unk_token": null
}

1
vocab.json Normal file
View File

File diff suppressed because one or more lines are too long