Datasets:

asingh15
/

glm52-aligned-rubric-traces

uid stringclasses 149 values	index int64 0 23.8k	writer_messages listlengths 1 1	reasoning stringlengths 89 79.6k	rubric_raw stringlengths 771 14.3k	rubric stringlengths 752 12.1k	finish_reason stringclasses 1 value	teacher stringclasses 1 value
AE3GITZC2MNOBOS72JZ35A345SKQ	16	[ { "role": "user", "content": "You are an expert at characterising how a specific human writes online book reviews.\n\nYou will be shown 8 examples of reviews written by ONE user, each preceded by the book's metadata. Your job is to write a rubric that captures this user's distinctive style, voice, content f...	Let me analyze this user's review style, preferences, and what they would likely say about "All You Could Ask For" by Mike Greenberg. Key observations about the user: 1. They give precise star ratings (often with decimals like 3.5, 3.75, 4.25, 4.75) 2. Reviews are concise (2-4 sentences, 15-70 words) 3. They use enthu...	<rubric> 1. Includes a precise star rating with decimal points (e.g., "4 out of 5 stars," "4.25 out of 5 stars," "4.5 out of 5 stars") placed at or near the beginning of the review, consistent with the user's established rating format across nearly all reviews. 2. Expresses an overall positive stance on the book (like...	1. Includes a precise star rating with decimal points (e.g., "4 out of 5 stars," "4.25 out of 5 stars," "4.5 out of 5 stars") placed at or near the beginning of the review, consistent with the user's established rating format across nearly all reviews. 2. Expresses an overall positive stance on the book (likely 4-5 st...	stop	glm52b
AE2WCYZ5MUGARAWQCCJY7NC7W4FQ	0	[ { "role": "user", "content": "You are an expert at characterising how a specific human writes online book reviews.\n\nYou will be shown 8 examples of reviews written by ONE user, each preceded by the book's metadata. Your job is to write a rubric that captures this user's distinctive style, voice, content f...	Let me analyze the user's review patterns and predict what they would say about "Natural Born Charmer" by Susan Elizabeth Phillips. Key observations about this user: 1. They primarily read paranormal romance, fantasy, and some science fiction 2. They like series books and follow character arcs 3. They value entertainm...	<rubric> 1. The review expresses an overall positive verdict on the book, consistent with this user's documented pattern of enjoying books outside their usual genre (as with "Illegal Alien") when they impulse-buy or try something new, while still noting it isn't their typical read. 2. The review specifically praises t...	1. The review expresses an overall positive verdict on the book, consistent with this user's documented pattern of enjoying books outside their usual genre (as with "Illegal Alien") when they impulse-buy or try something new, while still noting it isn't their typical read. 2. The review specifically praises the humor ...	stop	glm52b
AE7SJ5RH4P6BTE244I7CLPAQDGFQ	136	[ { "role": "user", "content": "You are an expert at characterising how a specific human writes online book reviews.\n\nYou will be shown 8 examples of reviews written by ONE user, each preceded by the book's metadata. Your job is to write a rubric that captures this user's distinctive style, voice, content f...	Let me analyze this user's reviewing style and predict what they would say about "Dark Awakening" by Kendra Leigh Castle. Book Context: - Paranormal romance about shapeshifting vampire Tynan MacGillivray - He's sent by his queen to find a Seer (human woman with special gift) - Lily Quinn is the Seer - has nightmar...	<rubric> 1. Overall verdict is strongly positive, framing Dark Awakening as an impressive paranormal romance debut/series starter with standout world-building — the user typically loves genre fiction that brings something fresh, and the vampire dynasty concept would excite her. 2. Specifically praises the vampire worl...	1. Overall verdict is strongly positive, framing Dark Awakening as an impressive paranormal romance debut/series starter with standout world-building — the user typically loves genre fiction that brings something fresh, and the vampire dynasty concept would excite her. 2. Specifically praises the vampire world-buildin...	stop	glm52b
AE7SJ5RH4P6BTE244I7CLPAQDGFQ	120	[ { "role": "user", "content": "You are an expert at characterising how a specific human writes online book reviews.\n\nYou will be shown 8 examples of reviews written by ONE user, each preceded by the book's metadata. Your job is to write a rubric that captures this user's distinctive style, voice, content f...	Let me analyze this user's review style and predict what they would say about "Dead on the Delta" by Stacey Jay. Book Analysis: - Urban fantasy set in Louisiana with fairies that are mutated and deadly (mosquito-size, blood-drinking) - Protagonist: Annabelle Lee - hard-drinking, smart-mouthed, bicycle-riding redhe...	<rubric> 1. Overall enthusiastic stance as a promising urban fantasy series debut. The candidate review should convey a positive verdict, framing Dead on the Delta as an exciting or strong start to a new series, consistent with this user's pattern of praising UF debuts (e.g., "A great urban fantasy debut!"). The to...	1. Overall enthusiastic stance as a promising urban fantasy series debut. The candidate review should convey a positive verdict, framing Dead on the Delta as an exciting or strong start to a new series, consistent with this user's pattern of praising UF debuts (e.g., "A great urban fantasy debut!"). The tone should...	stop	glm52b
AE3GITZC2MNOBOS72JZ35A345SKQ	32	[{"role":"user","content":"You are an expert at characterising how a specific human writes online bo(...TRUNCATED)	"Let me analyze the user's persona and past reviews carefully to predict what they would write about(...TRUNCATED)	"<rubric>\n1. Opens with or includes a numeric star rating (likely 4 to 4.5 out of 5, possibly with (...TRUNCATED)	"1. Opens with or includes a numeric star rating (likely 4 to 4.5 out of 5, possibly with a decimal (...TRUNCATED)	stop	glm52b
AE4OJYSCIEBY6JD2ZROWXLHTV3DQ	72	[{"role":"user","content":"You are an expert at characterising how a specific human writes online bo(...TRUNCATED)	"Let me analyze the user's persona and past reviews to predict what they would say about \"100 Words(...TRUNCATED)	"<rubric>\n1. Expresses an overall POSITIVE stance (4-5 stars) on the book, framed through the lens (...TRUNCATED)	"1. Expresses an overall POSITIVE stance (4-5 stars) on the book, framed through the lens of being a(...TRUNCATED)	stop	glm52b
AE7SJ5RH4P6BTE244I7CLPAQDGFQ	128	[{"role":"user","content":"You are an expert at characterising how a specific human writes online bo(...TRUNCATED)	"Let me analyze the user's review style and predict what they would say about \"Black Dust Mambo\" b(...TRUNCATED)	"<rubric>\n1. Review opens with an enthusiastic, exclamatory title or hook that signals a positive o(...TRUNCATED)	"1. Review opens with an enthusiastic, exclamatory title or hook that signals a positive overall ver(...TRUNCATED)	stop	glm52b
AE3GITZC2MNOBOS72JZ35A345SKQ	40	[{"role":"user","content":"You are an expert at characterising how a specific human writes online bo(...TRUNCATED)	"The user wants a rubric to evaluate a candidate review for the book \"The Secrets of Lake Road: A N(...TRUNCATED)	"<rubric>\n1. Awards a positive overall verdict with a specific, often precise star rating (e.g., \"(...TRUNCATED)	"1. Awards a positive overall verdict with a specific, often precise star rating (e.g., \"4.5 out of(...TRUNCATED)	stop	glm52b
AEAPBPDVH2ITSUQTUV3AZN5V3RXA	152	[{"role":"user","content":"You are an expert at characterising how a specific human writes online bo(...TRUNCATED)	"Let me analyze this user's review style, preferences, and patterns to predict what they would say a(...TRUNCATED)	"<rubric>\n1. Opens with a structured header identifying the book as a YA Mystery/Thriller, followed(...TRUNCATED)	"1. Opens with a structured header identifying the book as a YA Mystery/Thriller, followed by a rati(...TRUNCATED)	stop	glm52b
AE3GWRRBGX6JEHXHZ746D3AJWNCA	48	[{"role":"user","content":"You are an expert at characterising how a specific human writes online bo(...TRUNCATED)	"Let me analyze the user's past reviews and persona to predict what they would say about \"The Flyin(...TRUNCATED)	"<rubric>\n1. Overall stance is moderately positive but measured, praising the folktale's family-tog(...TRUNCATED)	"1. Overall stance is moderately positive but measured, praising the folktale's family-togetherness (...TRUNCATED)	stop	glm52b

End of preview. Expand in Data Studio

GLM-5.2 Aligned Rubric-Writing Traces

23777 teacher traces from GLM-5.2 on the aligned rubric-writing task, collected to distill / warmstart a smaller rubric-writer. For each (user, book) example the teacher is shown a persona-conditioned prompt (a user's past book reviews) and asked to (1) predict what that user would likely write about a new book and (2) produce a <rubric> of numbered criteria for scoring candidate reviews on coverage of that prediction. The full generation — thinking + rubric — is saved.

Fields

uid: reviewer id (groups traces by user)
index, sample: source row index and sample number (N=1: one generation per prompt)
writer_messages: the chat prompt sent to the teacher ([{"role":"user","content": ...}])
reasoning: the teacher's thinking (chain-of-thought)
rubric_raw: the teacher's final content (the <rubric>...</rubric> block)
rubric: the extracted rubric body
finish_reason: decoding stop reason (all stop)
teacher: teacher model tag (glm52b = GLM-5.2-FP8)

Generation

Training-matched decoding: temperature=1.0, top_p=1.0, thinking enabled, up to 32768 tokens.

Notes

The source-dataset gold (ground_truth: target reviews + candidate pools) is excluded. Traces are model outputs and may contain errors; intended for research (distillation/SFT).

Downloads last month: 44