jbuaba/iolai-2026-qwen25-14b
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: jbuaba/iolai-2026-qwen25-14b
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-09-13 04:46:18 +08:00
commit 13827db9e9
21 changed files with 457381 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
.gitattributes vendored Normal file
View File

@@ -0,0 +1,35 @@
*.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

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.

8
README.md Normal file
View File

@@ -0,0 +1,8 @@
# IOL-AI 2026 — final burn: ATB 0.178 + Lipas normalize + explain
Measured base: pad/hygiene/match + RP=1.0 + winner encode + explain 100%.
Add-only: Lipas `safe_normalize` for `match_letters` / `text_to_num`.
Expected: **~0.179–0.183** (peer ceiling 0.1828). Not a path to 0.19.
Nominate prior **0.1782 / explain 100%** if this regresses.

35
config.json Normal file
View File

@@ -0,0 +1,35 @@
{
"architectures": [
"Qwen2ForCausalLM"
],
"attention_dropout": 0.0,
"bos_token_id": 151643,
"eos_token_id": 151645,
"hidden_act": "silu",
"hidden_size": 5120,
"initializer_range": 0.02,
"intermediate_size": 13824,
"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_theta": 1000000.0,
"sliding_window": 131072,
"tie_word_embeddings": false,
"torch_dtype": "float16",
"transformers_version": "4.41.1",
"use_cache": true,
"use_sliding_window": false,
"vocab_size": 152064
}

11
generation_config.json Normal file
View File

@@ -0,0 +1,11 @@
{
"bos_token_id": 151643,
"do_sample": false,
"eos_token_id": [
151645,
151643
],
"pad_token_id": 151643,
"repetition_penalty": 1.0,
"transformers_version": "4.41.1"
}

151387
merges.txt Normal file
View File

File diff suppressed because it is too large Load Diff

3
model-00001-of-00003.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4e874dc3b1febb5b22fe74a8793066ae430d90cdbc51765d0a4eb44a82a1fbbd
size 3988804408

3
model-00002-of-00003.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b3b25da74cc854cdc726956f8152f1dda8519c7bb7d4724ac12c1312463e61c8
size 3968309440

3
model-00003-of-00003.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:25b97bf28033ed4560387293ce76bd3dc55a22882fea005a10c137b6478dae85
size 2023056736

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

File diff suppressed because it is too large Load Diff

163
script.py Normal file
View File

@@ -0,0 +1,163 @@
from __future__ import annotations
import json
import os
import time
os.environ["HF_HUB_OFFLINE"] = "1"
os.environ["TRANSFORMERS_OFFLINE"] = "1"
os.environ["TOKENIZERS_PARALLELISM"] = "false"
import pandas as pd
from solver.minimal import MAX_NEW_TOKENS, solve_row
from solver.model import DEFAULT_MODEL_ID, apply_greedy_decoding, assert_gpu_resident, load_model
MODEL_ID = os.environ.get("IOL_MODEL_ID", DEFAULT_MODEL_ID)
WRITE_EXPLANATIONS = os.environ.get("IOL_EXPLAIN", "1").strip().lower() in {
"1",
"true",
"yes",
}
TEST_CSV_PATH = os.environ.get("IOL_TEST_CSV", "/tmp/data/test.csv")
SUBMISSION_CSV_PATH = os.environ.get("IOL_OUT_CSV", "submission.csv")
HARD_LIMIT_SEC = float(os.environ.get("IOL_HARD_LIMIT_SEC", str(30 * 60)))
SAFETY_MARGIN_SEC = float(os.environ.get("IOL_SAFETY_SEC", "150"))
EXPLAIN_RESERVE_SEC = float(os.environ.get("IOL_EXPLAIN_RESERVE_SEC", "300"))
EXPLAIN_MAX_NEW_TOKENS = 96
def write_submission(rows: list[dict], path: str) -> None:
pd.DataFrame(rows).to_csv(path, index=False)
def submission_row(row, answers: list[str], explanation: str | None) -> dict:
if not answers:
answers = ["?"]
record = {"id": row["id"], "pred": json.dumps(answers, ensure_ascii=False)}
if WRITE_EXPLANATIONS:
record["explanation"] = explanation or ""
return record
def is_placeholder_prediction(pred_json: str) -> bool:
try:
return json.loads(pred_json) == ["?"]
except Exception:
return False
def main() -> None:
started = time.monotonic()
problems = pd.read_csv(TEST_CSV_PATH, dtype=str).fillna("")
submission = [submission_row(row, ["?"], "") for _, row in problems.iterrows()]
write_submission(submission, SUBMISSION_CSV_PATH)
print(f"placeholder {SUBMISSION_CSV_PATH} ({len(submission)} rows)", flush=True)
bundle = load_model(MODEL_ID, offline=True)
assert_gpu_resident(bundle)
apply_greedy_decoding(bundle.model)
try:
repetition_penalty = float(bundle.model.generation_config.repetition_penalty)
except Exception:
repetition_penalty = float("nan")
print(f"generation_config.repetition_penalty={repetition_penalty}", flush=True)
hard_stop = started + HARD_LIMIT_SEC - SAFETY_MARGIN_SEC
answer_deadline = hard_stop - (EXPLAIN_RESERVE_SEC if WRITE_EXPLANATIONS else 0.0)
print(
f"loaded {bundle.model_id} in {time.monotonic() - started:.1f}s "
f"explain={WRITE_EXPLANATIONS} answer_deadline_reserve="
f"{EXPLAIN_RESERVE_SEC if WRITE_EXPLANATIONS else 0:.0f}s",
flush=True,
)
rows_completed = 0
hit_token_cap = 0
soft_deadline_row: int | None = None
completed: list[tuple[object, list[str], str]] = []
for row_index, row in problems.iterrows():
if time.monotonic() >= answer_deadline:
soft_deadline_row = int(row_index)
print(f"soft deadline at {row_index}/{len(problems)}", flush=True)
break
answers, raw_text, stats = solve_row(
row, bundle, max_new_tokens=MAX_NEW_TOKENS
)
hit_token_cap += int(stats.hit_max_new)
submission[row_index] = submission_row(row, answers, None)
completed.append((row_index, answers, raw_text))
rows_completed += 1
write_submission(submission, SUBMISSION_CSV_PATH)
print(
f"{row_index + 1}/{len(problems)} n={len(answers)} "
f"new={stats.new_tokens} "
f"{'HIT_CAP' if stats.hit_max_new else 'eos'} "
f"elapsed={time.monotonic() - started:.0f}s",
flush=True,
)
explanations_written = 0
if WRITE_EXPLANATIONS:
for row_index, answers, raw_text in completed:
if time.monotonic() >= hard_stop:
break
explanation = write_explanation(
bundle, problems.loc[row_index], answers, raw_text
)
if explanation.strip():
explanations_written += 1
submission[row_index] = submission_row(
problems.loc[row_index], answers, explanation
)
write_submission(submission, SUBMISSION_CSV_PATH)
write_submission(submission, SUBMISSION_CSV_PATH)
placeholder_rows = sum(
1 for record in submission if is_placeholder_prediction(record["pred"])
)
explanation_rate = explanations_written / max(1, len(problems))
print(
f"summary n_rows={len(problems)} n_done={rows_completed} "
f"n_qmark={placeholder_rows} hit_cap={hit_token_cap} "
f"soft_deadline_at={soft_deadline_row} rp={repetition_penalty} "
f"explanation_rate={explanation_rate:.3f} "
f"total={time.monotonic() - started:.0f}s",
flush=True,
)
print(
f"wrote {SUBMISSION_CSV_PATH} total={time.monotonic() - started:.0f}s",
flush=True,
)
def write_explanation(bundle, row, answers: list[str], raw_text: str) -> str:
from solver.model import generate
messages = [
{
"role": "system",
"content": (
"Write 2-4 short bullet points explaining the answer. "
"Human-readable, not a chain of thought."
),
},
{
"role": "user",
"content": (
f"CONTEXT:\n{row.get('context', '')}\n\n"
f"QUERY:\n{row.get('query', '')}\n\n"
f"ANSWERS:\n{answers}\n\n"
f"TRACE:\n{(raw_text or '')[:2000]}"
),
},
]
try:
return generate(bundle, messages, max_new_tokens=EXPLAIN_MAX_NEW_TOKENS)
except Exception:
return ""
if __name__ == "__main__":
main()

26
solver/__init__.py Normal file
View File

@@ -0,0 +1,26 @@
from .items import count_answer_slots, pad_short_answers
from .matching import solve_matching
from .minimal import solve_row
from .model import (
GenStats,
ModelBundle,
apply_greedy_decoding,
generate,
generate_with_stats,
load_model,
)
from .normalize import safe_normalize_answers
__all__ = [
"GenStats",
"ModelBundle",
"apply_greedy_decoding",
"count_answer_slots",
"generate",
"generate_with_stats",
"load_model",
"pad_short_answers",
"safe_normalize_answers",
"solve_matching",
"solve_row",
]

96
solver/items.py Normal file
View File

@@ -0,0 +1,96 @@
from __future__ import annotations
import re
_LINE_NUMBER = re.compile(r"^[ \t]*(\d{1,3})[.)\]]", re.M)
_PAREN_NUMBER = re.compile(r"\((\d{1,3})\)")
_ITEM_RANGE = re.compile(r"\(?(\d{1,3})\s*(?:[-–—]|to)\s*(\d{1,3})\)?")
_LINE_LETTER = re.compile(r"^[ \t]*([A-Z])[.)\]]\s", re.M)
_PAREN_LETTER = re.compile(r"\(([A-Z])\)")
_NUMBERED_ANSWER = re.compile(r"^\s*\(?(\d{1,3})\)?[.):\]]\s*(.+)$")
def count_answer_slots(query: str, context: str = "") -> int:
"""How many answers this problem expects. Always >= 1."""
query = query or ""
line_nums = [int(m) for m in _LINE_NUMBER.findall(query)]
paren_nums = [int(m) for m in _PAREN_NUMBER.findall(query)]
range_count = 0
for start, end in _ITEM_RANGE.findall(query):
start_i, end_i = int(start), int(end)
if 0 < end_i - start_i < 60:
range_count = max(range_count, end_i - start_i + 1)
marker_count = max(len(set(line_nums)), len(set(paren_nums)))
if range_count and marker_count and range_count != marker_count:
return marker_count
marker_count = max(
marker_count,
len(set(_LINE_LETTER.findall(query))),
len(set(_PAREN_LETTER.findall(query))),
)
slot_count = max(range_count, marker_count)
if slot_count > 1:
return slot_count
lines = [line.strip() for line in query.splitlines() if line.strip()]
if len(lines) > 1:
head = lines[0]
body = lines[1:] if head.endswith((":", ".")) else lines
if body:
return len(body)
if context:
context_nums = len(set(int(m) for m in _LINE_NUMBER.findall(context)))
if context_nums > 1:
return context_nums
context_letters = len(set(_LINE_LETTER.findall(context)))
if context_letters > 1:
return context_letters
return max(slot_count, 1)
def source_fallbacks(query: str, slot_count: int) -> list[str]:
"""Per-item source text used when the model returns too few lines.
Empty predictions score zero on EM and chrF. Echoing the query item is
usually wrong on EM but recovers chrF on fill-blank / transcription tasks.
"""
query = query or ""
sources: list[str] = []
for line in query.splitlines():
stripped = line.strip()
if not stripped:
continue
match = _NUMBERED_ANSWER.match(stripped)
if match:
sources.append(match.group(2).strip())
if not sources:
lines = [line.strip() for line in query.splitlines() if line.strip()]
if len(lines) > 1 and lines[0].endswith((":", ".")):
sources = lines[1:]
sources = [s.split("|")[0].strip() if "|" in s else s for s in sources]
sources = [s for s in sources if s]
while len(sources) < slot_count:
sources.append(sources[-1] if sources else "?")
return sources[:slot_count]
def pad_short_answers(
answers: list[str],
slot_count: int,
fallbacks: list[str] | None = None,
) -> list[str]:
"""Pad undersized answer lists only. Never truncate — the grader keeps the first N."""
slot_count = max(1, int(slot_count))
padded = [str(a).strip() if a and str(a).strip() else "?" for a in answers]
while len(padded) < slot_count:
if fallbacks and len(padded) < len(fallbacks):
fill = str(fallbacks[len(padded)]).strip() or "?"
else:
fill = "?"
padded.append(fill)
return padded

173
solver/matching.py Normal file
View File

@@ -0,0 +1,173 @@
from __future__ import annotations
import re
from typing import Any
from .model import ModelBundle
_OPT_LINE = re.compile(r"^[ \t]*([A-Za-z])[.)]\s+(.+)$", re.M)
_ITEM_LINE = re.compile(r"^[ \t]*(\d{1,3})[.)]\s+(.+)$", re.M)
def parse_matching_block(
context: str,
) -> tuple[list[tuple[int, str]], list[tuple[str, str]]]:
items = [(int(a), b.strip()) for a, b in _ITEM_LINE.findall(context or "")]
opts = [(a.upper(), b.strip()) for a, b in _OPT_LINE.findall(context or "")]
seen_items: set[int] = set()
items = [x for x in items if not (x[0] in seen_items or seen_items.add(x[0]))]
seen_opts: set[str] = set()
opts = [x for x in opts if not (x[0] in seen_opts or seen_opts.add(x[0]))]
return items, opts
def best_assignment(score: list[list[float]]) -> list[int]:
n = len(score)
m = len(score[0]) if score else 0
if n == 0 or m == 0:
return []
try:
import numpy as np
from scipy.optimize import linear_sum_assignment
_, cols = linear_sum_assignment(-np.array(score, dtype=float))
return list(cols)
except Exception:
pass
used: set[int] = set()
out = [0] * n
order = sorted(
range(n),
key=lambda i: -(
max(score[i]) - sorted(score[i])[-2] if m > 1 else max(score[i])
),
)
for i in order:
j = max(
(jj for jj in range(m) if jj not in used),
key=lambda jj: score[i][jj],
default=0,
)
used.add(j)
out[i] = j
for _ in range(4):
improved = False
for a in range(n):
for b in range(a + 1, n):
cur = score[a][out[a]] + score[b][out[b]]
alt = score[a][out[b]] + score[b][out[a]]
if alt > cur + 1e-9:
out[a], out[b] = out[b], out[a]
improved = True
if not improved:
break
return out
def repair_letter_bijection(answers: list[str]) -> list[str]:
if len(answers) < 3:
return answers
if not all(re.fullmatch(r"[A-Z]", a or "") for a in answers):
return answers
n = len(answers)
universe = [chr(ord("A") + i) for i in range(n)]
if len(set(answers)) == n:
return answers
unused = [letter for letter in universe if letter not in set(answers)]
if not unused:
return answers
seen: set[str] = set()
out: list[str] = []
for answer in answers:
if answer in seen and unused:
out.append(unused.pop(0))
else:
seen.add(answer)
out.append(answer)
return out
def solve_matching(
bundle: ModelBundle,
row: Any,
slot_count: int,
*,
batch_size: int = 4,
) -> list[str] | None:
"""Score (item, option) next-token logprobs and take a 1-1 assignment.
Returns None on any failure so callers can fall back to greedy decode.
"""
import torch
items, opts = parse_matching_block(str(row.get("context", "") or ""))
if len(items) < 3 or len(opts) < 3 or len(items) != slot_count:
return None
letters = [opt[0] for opt in opts]
candidate_ids: list[list[int]] = []
for letter in letters:
ids: set[int] = set()
for form in (letter, " " + letter):
tokens = bundle.tok.encode(form, add_special_tokens=False)
if tokens:
ids.add(int(tokens[0]))
if not ids:
return None
candidate_ids.append(sorted(ids))
context = str(row.get("context", "") or "").strip()
prompts: list[str] = []
for number, item_text in items:
messages = [
{
"role": "system",
"content": (
"You match items to their correct counterparts in a "
"linguistics problem. Reply with one option letter only."
),
},
{
"role": "user",
"content": (
f"{context}\n\nWhich lettered option corresponds to item "
f"{number} ({item_text})? Reply with the option letter only."
),
},
]
prompts.append(
bundle.tok.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
)
score: list[list[float]] = []
try:
for start in range(0, len(prompts), batch_size):
chunk = prompts[start : start + batch_size]
encoded = bundle.tok(
chunk,
return_tensors="pt",
padding=True,
truncation=True,
max_length=6144,
)
encoded = {k: v.to(bundle.model.device) for k, v in encoded.items()}
with torch.no_grad():
logits = bundle.model(**encoded).logits[:, -1, :].float()
logprobs = torch.log_softmax(logits, dim=-1)
for batch_index in range(len(chunk)):
score.append(
[
max(float(logprobs[batch_index, token_id].item()) for token_id in ids)
for ids in candidate_ids
]
)
except Exception:
return None
columns = best_assignment(score)
if len(columns) != slot_count:
return None
return repair_letter_bijection([letters[col] for col in columns])

145
solver/minimal.py Normal file
View File

@@ -0,0 +1,145 @@
from __future__ import annotations
import re
from typing import Any, Mapping
from .items import count_answer_slots, pad_short_answers, source_fallbacks
from .matching import solve_matching
from .model import GenStats, ModelBundle, generate_with_stats
from .normalize import safe_normalize_answers
SYSTEM_PROMPT = (
"You solve International Linguistics Olympiad problems. "
"Answer every numbered item. Put each answer on its own line, "
"in order, with no numbering and no extra text."
)
MAX_NEW_TOKENS = 512
_LEADING_MARKER = re.compile(r"^\s*(?:\(?\d{1,3}\)?[.):\]]\s*|[-*•]\s+)")
_CODE_FENCE = re.compile(r"^```[a-zA-Z]*\s*$")
_PREAMBLE = re.compile(
r"^\s*(?:here (?:are|is)\b|answers?\s*:?\s*$|explanation\b|note\b|okay\b|"
r"solution\b|reasoning\b|analysis\b|translations?\s*:?\s*$|the answers?\b|"
r"let me\b|first,|so,|therefore\b|thus\b)",
re.I,
)
_NUMBERED_LINE = re.compile(r"^\s*\(?(\d{1,3})\)?[.):\]]\s*(.+)$")
def clean_answer_text(text: str) -> str:
text = (text or "").strip()
text = _LEADING_MARKER.sub("", text)
text = text.strip().strip("`").strip()
if len(text) >= 2 and text[0] == text[-1] and text[0] in "\"'“”":
text = text[1:-1].strip()
return text.strip()
def extract_answer_lines(model_text: str, slot_count: int) -> list[str]:
labeled: dict[int, str] = {}
unlabeled: list[str] = []
for line in (model_text or "").splitlines():
if not line.strip() or _CODE_FENCE.match(line):
continue
if _PREAMBLE.match(line):
continue
numbered = _NUMBERED_LINE.match(line.strip())
if numbered:
label = int(numbered.group(1))
value = clean_answer_text(numbered.group(2))
if value and not _PREAMBLE.match(value):
labeled[label] = value
continue
cleaned = clean_answer_text(line)
if cleaned and not _PREAMBLE.match(cleaned):
unlabeled.append(cleaned)
if labeled:
slots: list[str | None] = [None] * slot_count
for label, value in labeled.items():
if 1 <= label <= slot_count:
slots[label - 1] = value
fill_from = 0
for index in range(slot_count):
if slots[index] is None and fill_from < len(unlabeled):
slots[index] = unlabeled[fill_from]
fill_from += 1
ordered = [s for s in slots if s is not None]
leftover = unlabeled[fill_from:]
return ordered + leftover
return unlabeled
def _greedy_answers(
row: Mapping[str, Any],
bundle: ModelBundle,
*,
max_new_tokens: int,
generate_fn,
) -> tuple[list[str], str, GenStats]:
context = str(row.get("context", "") or "").strip()
query = str(row.get("query", "") or "").strip()
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": f"{context}\n\n{query}"},
]
if generate_fn is not None:
raw = generate_fn(bundle, messages, max_new_tokens)
stats = GenStats(
prompt_tokens=0,
new_tokens=0,
hit_max_new=False,
eos_limited=True,
)
else:
raw, stats = generate_with_stats(
bundle, messages, max_new_tokens=max_new_tokens
)
slot_count = count_answer_slots(query, context)
fallbacks = source_fallbacks(query, slot_count)
answers = pad_short_answers(
extract_answer_lines(raw, slot_count),
slot_count,
fallbacks,
)
return answers, raw, stats
def solve_row(
row: Mapping[str, Any],
bundle: ModelBundle,
*,
max_new_tokens: int = MAX_NEW_TOKENS,
generate_fn=None,
) -> tuple[list[str], str, GenStats]:
context = str(row.get("context", "") or "").strip()
query = str(row.get("query", "") or "").strip()
task_type = str(row.get("task_type", "") or "").strip().lower()
slot_count = count_answer_slots(query, context)
if task_type == "match_letters" and generate_fn is None:
try:
matched = solve_matching(bundle, row, slot_count)
if matched and len(matched) == slot_count:
stats = GenStats(
prompt_tokens=0,
new_tokens=0,
hit_max_new=False,
eos_limited=True,
)
return (
safe_normalize_answers(matched, task_type),
"",
stats,
)
except Exception:
pass
answers, raw, stats = _greedy_answers(
row, bundle, max_new_tokens=max_new_tokens, generate_fn=generate_fn
)
return safe_normalize_answers(answers, task_type), raw, stats

202
solver/model.py Normal file
View File

@@ -0,0 +1,202 @@
from __future__ import annotations
import os
from dataclasses import dataclass
from typing import Any
DEFAULT_MODEL_ID = "."
LOAD_MODE_AWQ = "awq"
LOAD_MODE_BNB = "bnb"
@dataclass
class ModelBundle:
tok: Any
model: Any
model_id: str
@dataclass(frozen=True)
class GenStats:
prompt_tokens: int
new_tokens: int
hit_max_new: bool
eos_limited: bool
def assert_gpu_resident(bundle: ModelBundle) -> None:
import torch
if not torch.cuda.is_available():
print("warn: CUDA unavailable", flush=True)
return
bad = []
for name, param in bundle.model.named_parameters():
if not str(param.device).startswith("cuda"):
bad.append(f"{name}:{param.device}")
if len(bad) >= 5:
break
if bad:
print(f"warn: non-CUDA parameters: {bad}", flush=True)
return
print(
f"gpu ok | VRAM {torch.cuda.memory_allocated() / 1e9:.2f} GB",
flush=True,
)
def load_model(
model_id: str | None = None,
*,
offline: bool | None = None,
load_mode: str | None = None,
) -> ModelBundle:
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = model_id or os.environ.get("IOL_MODEL_ID", DEFAULT_MODEL_ID)
load_mode = (load_mode or os.environ.get("IOL_LOAD", LOAD_MODE_AWQ)).strip().lower()
if offline is None:
offline = model_id == DEFAULT_MODEL_ID or os.environ.get("HF_HUB_OFFLINE") == "1"
if offline:
os.environ["HF_HUB_OFFLINE"] = "1"
os.environ["TRANSFORMERS_OFFLINE"] = "1"
else:
os.environ.pop("HF_HUB_OFFLINE", None)
os.environ.pop("TRANSFORMERS_OFFLINE", None)
os.environ.setdefault("TOKENIZERS_PARALLELISM", "false")
tok = AutoTokenizer.from_pretrained(model_id)
if tok.pad_token_id is None and tok.eos_token_id is not None:
tok.pad_token = tok.eos_token
dtype_kwargs = _dtype_kwargs(torch)
preferred: Any = {"": 0} if torch.cuda.is_available() else "auto"
def _load(device_map: Any):
if load_mode == LOAD_MODE_BNB:
from transformers import BitsAndBytesConfig
return AutoModelForCausalLM.from_pretrained(
model_id,
quantization_config=BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_compute_dtype=torch.float16,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
),
device_map=device_map,
).eval()
try:
return AutoModelForCausalLM.from_pretrained(
model_id,
device_map=device_map,
**dtype_kwargs,
).eval()
except ImportError as exc:
raise ImportError(
"AWQ load failed; install gptqmodel/autoawq or use IOL_LOAD=bnb"
) from exc
try:
model = _load(preferred)
except ImportError:
raise
except Exception as exc:
if preferred == "auto":
raise
print(f"warn: device_map retry auto ({exc})", flush=True)
model = _load("auto")
apply_greedy_decoding(model)
return ModelBundle(tok=tok, model=model, model_id=model_id)
def apply_greedy_decoding(model: Any) -> None:
try:
cfg = model.generation_config
cfg.do_sample = False
cfg.repetition_penalty = 1.0
for key in ("temperature", "top_p", "top_k", "typical_p"):
if hasattr(cfg, key):
setattr(cfg, key, None)
except Exception:
pass
def _dtype_kwargs(torch_mod) -> dict:
try:
import inspect
from transformers import AutoModelForCausalLM
if "dtype" in inspect.signature(AutoModelForCausalLM.from_pretrained).parameters:
return {"dtype": torch_mod.float16}
except Exception:
pass
return {"torch_dtype": torch_mod.float16}
def _prompt_tensors(tok: Any, model: Any, messages: list[dict[str, str]]):
text = tok.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
)
enc = tok(
text,
return_tensors="pt",
truncation=True,
max_length=6144,
)
moved = {k: v.to(model.device) for k, v in enc.items()}
return moved, int(moved["input_ids"].shape[-1])
def _pad_token_id(bundle: ModelBundle) -> int | None:
if getattr(bundle.tok, "pad_token_id", None) is not None:
return int(bundle.tok.pad_token_id)
if getattr(bundle.tok, "eos_token_id", None) is not None:
return int(bundle.tok.eos_token_id)
return None
def generate(
bundle: ModelBundle,
messages: list[dict[str, str]],
*,
max_new_tokens: int = 512,
) -> str:
text, _ = generate_with_stats(bundle, messages, max_new_tokens=max_new_tokens)
return text
def generate_with_stats(
bundle: ModelBundle,
messages: list[dict[str, str]],
*,
max_new_tokens: int = 512,
) -> tuple[str, GenStats]:
import torch
apply_greedy_decoding(bundle.model)
inputs, prompt_len = _prompt_tensors(bundle.tok, bundle.model, messages)
kwargs: dict[str, Any] = {
"max_new_tokens": max_new_tokens,
"do_sample": False,
"repetition_penalty": 1.0,
}
pad_id = _pad_token_id(bundle)
if pad_id is not None:
kwargs["pad_token_id"] = pad_id
with torch.no_grad():
output = bundle.model.generate(**inputs, **kwargs)
new_tokens = int(output.shape[-1] - prompt_len)
hit_max = new_tokens >= max_new_tokens
text = bundle.tok.decode(output[0][prompt_len:], skip_special_tokens=True).strip()
return text, GenStats(
prompt_tokens=int(prompt_len),
new_tokens=new_tokens,
hit_max_new=hit_max,
eos_limited=not hit_max,
)

42
solver/normalize.py Normal file
View File

@@ -0,0 +1,42 @@
"""Per-line surface normalizers (Lipas/Hul). No arity force / pad / truncate."""
from __future__ import annotations
import re
def normalize_match_letter(ans: str) -> str:
ans = ans.strip()
m = re.fullmatch(r"[\(\[]?([A-Za-z])[\)\]]?[.)]?", ans)
if m:
return m.group(1).upper()
tokens = re.findall(r"\b([A-Za-z])\b", ans)
if tokens:
return tokens[-1].upper()
m = re.search(r"[A-Za-z]", ans)
return m.group(0).upper() if m else ans
def normalize_text_to_num(ans: str) -> str:
a = re.sub(r"(?i)^(answer|ans|result)\s*[:=]\s*", "", ans.strip()).strip()
if re.fullmatch(r"[\d\s+\-*/^=()]+", a.replace(",", "")):
a = a.replace(",", "").replace(" ", "")
if "=" in a and " = " not in a:
a = a.replace("=", " = ")
return a.strip()
m = re.search(r"\d+", a)
return m.group(0) if m and len(a) < 40 else a
def safe_normalize_answers(answers: list[str], task_type: str) -> list[str]:
"""Per-line only. Does not pad, truncate, or reorder."""
task_type = (task_type or "").strip().lower()
out: list[str] = []
for a in answers:
a = a.strip()
if task_type == "match_letters":
a = normalize_match_letter(a)
elif task_type == "text_to_num":
a = normalize_text_to_num(a)
out.append(a)
return out

99
solver/runtime.py Normal file
View File

@@ -0,0 +1,99 @@
"""Offline runtime bootstrap for Qwen3 under the Space's older transformers."""
from __future__ import annotations
import importlib
import importlib.metadata
import os
import subprocess
import sys
from pathlib import Path
RUNTIME_PACKAGES = {
"transformers": "4.51.3",
"tokenizers": "0.21.1",
"huggingface_hub": "0.30.2",
"autoawq": "0.2.9",
}
RUNTIME_WHEELS = (
"transformers-4.51.3-py3-none-any.whl",
"tokenizers-0.21.1-cp39-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl",
"huggingface_hub-0.30.2-py3-none-any.whl",
"autoawq-0.2.9-py3-none-any.whl",
)
def _wheelhouse() -> Path:
env = os.environ.get("IOL_WHEELHOUSE") or os.environ.get("QWEN3_WHEELHOUSE")
if env:
return Path(env)
return Path(__file__).resolve().parent.parent / "wheelhouse"
def _runtime_dir() -> Path:
return Path(os.environ.get("IOL_RUNTIME_DIR", "/tmp/iol_qwen3_runtime"))
def installed_versions() -> dict[str, str]:
versions: dict[str, str] = {}
for package in RUNTIME_PACKAGES:
try:
versions[package] = importlib.metadata.version(package)
except importlib.metadata.PackageNotFoundError:
versions[package] = "missing"
return versions
def ensure_runtime() -> dict[str, str]:
"""Install bundled wheels into /tmp and prefer them on sys.path.
No-op when wheelhouse is absent (local Qwen2.5 packs / unit tests).
"""
wheelhouse = _wheelhouse()
if not wheelhouse.is_dir():
return installed_versions()
wheel_paths = [wheelhouse / name for name in RUNTIME_WHEELS]
missing = [str(path) for path in wheel_paths if not path.is_file()]
if missing:
raise FileNotFoundError(f"Missing offline runtime wheels: {missing}")
runtime_dir = _runtime_dir()
marker = runtime_dir / ".iol-qwen3-runtime-v1"
if not marker.is_file():
runtime_dir.mkdir(parents=True, exist_ok=True)
subprocess.run(
[
sys.executable,
"-m",
"pip",
"install",
"--disable-pip-version-check",
"--no-index",
"--no-deps",
"--upgrade",
"--target",
str(runtime_dir),
*(str(path) for path in wheel_paths),
],
check=True,
timeout=180,
)
marker.write_text("offline Qwen3 runtime installed\n", encoding="utf-8")
runtime_path = str(runtime_dir)
if runtime_path in sys.path:
sys.path.remove(runtime_path)
sys.path.insert(0, runtime_path)
importlib.invalidate_caches()
versions = installed_versions()
mismatches = {
name: (versions[name], expected)
for name, expected in RUNTIME_PACKAGES.items()
if versions[name] != expected
}
if mismatches:
raise RuntimeError(f"Offline runtime version mismatch: {mismatches}")
print(f"offline runtime: {versions}", flush=True)
return versions

303282
tokenizer.json Normal file
View File

File diff suppressed because it is too large Load Diff

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