\u672c\u6b21\u89c2\u5bdf\u8986\u76d6 Australia\/Sydney \u65f6\u95f4 2026\u5e7411\u67088\u65e5 07:55 \u81f3 11\u67089\u65e5 07:55\uff0c\u5b9e\u9645\u6536\u5f55 1 \u9879\u7ecf\u6838\u9a8c\u7684\u6280\u672f\u53d8\u5316\u3002<\/p>\n
TwIL-8B \u4ece\u5ba3\u5e03\u63a8\u8fdb\u5230\u516c\u5f00\u6743\u91cd<\/p>\n
webAI \u4e8e 11\u67088\u65e5\u5728 Hugging Face \u5efa\u7acb TwIL-8B \u4ed3\u5e93\u5e76\u901a\u8fc7\u9996\u4e2a\u63d0\u4ea4\u516c\u5f00\u5b8c\u6574\u4ea4\u4ed8\u7269\u3002\u6b64\u524d\u53ef\u6838\u9a8c\u7684\u516c\u5f00\u72b6\u6001\u662f 11\u67084\u65e5\u7684\u6a21\u578b\u5ba3\u5e03\uff1b\u672c\u6b21\u65b0\u589e\u4e8b\u5b9e\u662f\u4ed3\u5e93\u5b9e\u9645\u51fa\u73b0\u6a21\u578b\u5361\u3001\u8bb8\u53ef\u6587\u4ef6\u3001\u914d\u7f6e\u3001\u5206\u8bcd\u5668\u3001\u5e73\u8861\u7248\u4e0e\u8be6\u7ec6\u63a8\u7406\u7248\u7684 safetensors \u6743\u91cd\uff0c\u4ee5\u53ca\u591a\u79cd GGUF \u91cf\u5316\u6587\u4ef6\u3002\u4ed3\u5e93\u63d0\u4ea4\u5386\u53f2\u53ea\u6709\u4e00\u4e2a\u6839\u63d0\u4ea4\uff0c\u6587\u4ef6\u6811\u53ef\u76f4\u63a5\u6838\u5bf9\u6743\u91cd\u548c\u91cf\u5316\u6587\u4ef6\uff0c\u56e0\u6b64\u201c\u6743\u91cd\u5df2\u7ecf\u516c\u5f00\u201d\u5f97\u5230\u539f\u59cb\u8d44\u6599\u786e\u8ba4\u3002<\/p>\n
\u6a21\u578b\u5361\u5c06 TwIL-8B \u63cf\u8ff0\u4e3a\u57fa\u4e8e IBM Granite 4.2 8B \u540e\u8bad\u7ec3\u7684 87.9 \u4ebf\u53c2\u6570\u7a20\u5bc6\u6587\u672c\u6a21\u578b\uff0c\u8bad\u7ec3\u6d41\u7a0b\u5305\u62ec\u5168\u53c2\u6570\u76d1\u7763\u5fae\u8c03\u3001\u7528\u4e8e\u8be6\u7ec6\u63a8\u7406\u7248\u7684 GRPO\/DAPO \u5f3a\u5316\u5b66\u4e60\uff0c\u4ee5\u53ca\u7528\u4e8e\u5e73\u8861\u7248\u7684\u957f\u5ea6\u611f\u77e5\u5f3a\u5316\u5b66\u4e60\u3002\u4e24\u4e2a\u7248\u672c\u5171\u7528\u67b6\u6784\u3001\u5206\u8bcd\u5668\u548c\u804a\u5929\u6a21\u677f\uff1b\u9ed8\u8ba4\u4ed3\u5e93\u6839\u76ee\u5f55\u52a0\u8f7d\u5e73\u8861\u7248\uff0c\u8be6\u7ec6\u63a8\u7406\u7248\u4f4d\u4e8e verbose \u5b50\u76ee\u5f55\u3002\u5176\u4e0a\u4e0b\u6587\u7a97\u53e3\u4e3a\u7ee7\u627f\u81ea\u57fa\u7840\u6a21\u578b\u7684 131,072 \u4e2a token\uff0c\u8f93\u5165\u8f93\u51fa\u5747\u4e3a\u82f1\u6587\u6587\u672c\u3002\u6bcf\u4e2a bfloat16 \u7248\u672c\u7ea6 16.38 GiB\uff0c\u5e73\u8861\u7248 Q4_K_M GGUF \u7ea6 4.98 GiB\uff0c\u53ef\u7528\u4e8e\u672c\u5730\u63a8\u7406\u3002<\/p>\n
\u5f00\u653e\u6761\u4ef6\u5b58\u5728\u660e\u786e\u8fb9\u754c\u3002TwIL-8B \u4f7f\u7528 webAI Non-Commercial License 1.0\uff0c\u57fa\u7840\u6a21\u578b\u4ecd\u4fdd\u7559 IBM \u7684 Apache 2.0 \u8bb8\u53ef\uff1b\u56e0\u6b64\u8fd9\u662f\u53ef\u4e0b\u8f7d\u6743\u91cd\u53d1\u5e03\uff0c\u5e76\u975e\u5141\u8bb8\u5546\u4e1a\u4f7f\u7528\u7684\u5f00\u653e\u8bb8\u53ef\u3002\u6a21\u578b\u5361\u8fd8\u5199\u660e\uff0c\u9664 Granite 4.2 \u5df2\u6709\u63aa\u65bd\u5916\uff0c\u6ca1\u6709\u989d\u5916\u7684\u5b89\u5168\u6216\u504f\u597d\u8bad\u7ec3\uff0c\u5e76\u5c06\u77e5\u8bc6\u5bc6\u96c6\u578b\u4efb\u52a1\u3001\u4ec5\u652f\u6301\u82f1\u6587\u4ee5\u53ca\u8f83\u957f\u63a8\u7406\u8f93\u51fa\u5217\u4e3a\u9650\u5236\u3002<\/p>\n
\u5f00\u53d1\u8005\u62a5\u544a\u5e73\u8861\u7248\u5728 MATH-500 \u5f97\u5206 98.2\uff0c\u8be6\u7ec6\u63a8\u7406\u7248\u5728 LiveCodeBench \u5f97\u5206 81.6\u3002\u6a21\u578b\u5361\u660e\u786e\u8bf4\u660e\uff0c\u5e26\u6807\u8bb0\u7684\u90e8\u5206\u6307\u6807\u5c1a\u672a\u5728\u5e73\u8861\u7248\u8fd0\u884c\uff0c\u5bf9\u7167\u6a21\u578b\u6210\u7ee9\u6765\u81ea\u516c\u5f00\u8d44\u6599\u800c\u975e\u540c\u4e00\u56e2\u961f\u91cd\u65b0\u6d4b\u8bd5\u3002\u72ec\u7acb\u8ddf\u8e2a\u9875\u9762\u622a\u81f3\u672c\u6b21\u8fd0\u884c\u4ecd\u628a\u8fd9\u4e9b\u6210\u7ee9\u6807\u4e3a\u5f00\u53d1\u8005\u81ea\u62a5\uff0c\u5c1a\u65e0\u72ec\u7acb\u6392\u884c\u699c\u7ed3\u679c\u6216\u516c\u5f00\u590d\u73b0\u3002\u56e0\u6b64\uff0c\u672c\u6b21\u53ea\u8bb0\u5f55\u6743\u91cd\u4ea4\u4ed8\u3001\u6a21\u578b\u7ed3\u6784\u3001\u8bb8\u53ef\u548c\u5f00\u53d1\u8005\u62ab\u9732\u7684\u8bc4\u6d4b\u72b6\u6001\uff0c\u4e0d\u628a\u8fd9\u4e9b\u5206\u6570\u5199\u6210\u5df2\u83b7\u72ec\u7acb\u786e\u8ba4\u7684\u80fd\u529b\u7ed3\u8bba\u3002<\/p>\n
\u6765\u6e90<\/p>\n
webAI \/ Hugging Face\uff0cTwIL-8B model card\uff0c2026\u5e7411\u67088\u65e5\uff0chttps:\/\/huggingface.co\/webAI-Official\/TwIL-8B<\/p>\n
webAI \/ Hugging Face\uff0cPublish TwIL-8B commit\uff0c2026\u5e7411\u67088\u65e5\uff0chttps:\/\/huggingface.co\/webAI-Official\/TwIL-8B\/commit\/9563ba7dfb70eeca4a26654b195011cbf818e7d9<\/p>\n
isitsota\uff0cIs TwIL-8B state of the art? Pending benchmarks\uff0c2026\u5e7411\u67084\u65e5\uff0chttps:\/\/isitsota.com\/m\/twil-8b<\/p>\n
","protected":false},"excerpt":{"rendered":"","protected":false},"author":17897162,"featured_media":0,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_coblocks_attr":"","_coblocks_dimensions":"","_coblocks_responsive_height":"","_coblocks_accordion_ie_support":"","_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_geoff_meta_description_en":"webAI has publicly released TwIL-8B weights in balanced and verbose variants, along with quantised files, a model card, a non-commercial licence, developer evaluations and stated limitations.","_geoff_meta_description_zh":"webAI \u6b63\u5f0f\u516c\u5f00 TwIL-8B \u7684\u5e73\u8861\u7248\u4e0e\u8be6\u7ec6\u63a8\u7406\u7248\u6743\u91cd\u3001\u91cf\u5316\u6587\u4ef6\u3001\u6a21\u578b\u5361\u548c\u975e\u5546\u4e1a\u8bb8\u53ef\uff0c\u5e76\u62ab\u9732\u6a21\u578b\u7ed3\u6784\u3001\u8bad\u7ec3\u65b9\u5f0f\u3001\u5f00\u53d1\u8005\u8bc4\u6d4b\u53ca\u9650\u5236\u3002","_wpcom_ai_launchpad_first_post":false,"_jetpack_hide_featured_image":false,"_jetpack_featured_image_duplicate_dismissed":0,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2},"jetpack_post_was_ever_published":false},"categories":[787298266],"tags":[787296348,787298267,787296349,787298268,787294346,787294292],"class_list":["post-23041","post","type-post","status-publish","format-standard","hentry","category-ai-technology-observation","tag-ai-research","tag-ai-technology","tag-ai-research-zh","tag-ai-technology-zh","tag-artificial-intelligence","tag-787294292",""],"translatepress_titles":{"en_AU":"AI Technology Observations | 9 November 2026, 07:00","zh_CN":"AI\u6280\u672f\u89c2\u5bdf\uff5c2026\u5e7411\u67089\u65e5 07\u65f6"},"jetpack_publicize_connections":[],"jetpack_likes_enabled":false,"jetpack_sharing_enabled":true,"jetpack_shortlink":"https:\/\/wp.me\/p1ar9H-5ZD","jetpack_featured_media_url":"","_links":{"self":[{"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/posts\/23041","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/users\/17897162"}],"replies":[{"embeddable":true,"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/comments?post=23041"}],"version-history":[{"count":0,"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/posts\/23041\/revisions"}],"wp:attachment":[{"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/media?parent=23041"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/categories?post=23041"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/geoffreychen.com\/zh\/wp-json\/wp\/v2\/tags?post=23041"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}