llm-tuning-patterns

Name: llm-tuning-patterns
Author: parcadei

3.9kstars297forks

How to install llm-tuning-patterns?

parcadei/continuous-claude-v3/llm-tuning-patterns

$npx -y skills add parcadei/Continuous-Claude-v3 --skill llm-tuning-patterns

Installs into the current project.

Summary

LLM Tuning Patterns

Files · 1

View on GitHub

SKILL.md

1	# LLM Tuning Patterns
2
3	Evidence-based patterns for configuring LLM parameters, based on APOLLO and Godel-Prover research.
4
5	## Pattern
6
7	Different tasks require different LLM configurations. Use these evidence-based settings.
8
9	## Theorem Proving / Formal Reasoning
10
11	Based on APOLLO parity analysis:
12
13	\| Parameter \| Value \| Rationale \|
14	\|-----------\|-------\|-----------\|
15	\| max_tokens \| 4096 \| Proofs need space for chain-of-thought \|
16	\| temperature \| 0.6 \| Higher creativity for tactic exploration \|
17	\| top_p \| 0.95 \| Allow diverse proof paths \|
18
19	### Proof Plan Prompt
20
21	Always request a proof plan before tactics:
22
23	```
24	Given the theorem to prove:
25	[theorem statement]
26
27	First, write a high-level proof plan explaining your approach.
28	Then, suggest Lean 4 tactics to implement each step.
29	```
30
31	The proof plan (chain-of-thought) significantly improves tactic quality.
32
33	### Parallel Sampling
34
35	For hard proofs, use parallel sampling:
36	- Generate N=8-32 candidate proof attempts
37	- Use best-of-N selection
38	- Each sample at temperature 0.6-0.8
39
40	## Code Generation
41
42	\| Parameter \| Value \| Rationale \|
43	\|-----------\|-------\|-----------\|
44	\| max_tokens \| 2048 \| Sufficient for most functions \|
45	\| temperature \| 0.2-0.4 \| Prefer deterministic output \|
46
47	## Creative / Exploration Tasks
48
49	\| Parameter \| Value \| Rationale \|
50	\|-----------\|-------\|-----------\|
51	\| max_tokens \| 4096 \| Space for exploration \|
52	\| temperature \| 0.8-1.0 \| Maximum creativity \|
53
54	## Anti-Patterns
55
56	- Too low tokens for proofs: 512 tokens truncates chain-of-thought
57	- Too low temperature for proofs: 0.2 misses creative tactic paths
58	- No proof plan: Jumping to tactics without planning reduces success rate
59
60	## Source Sessions
61
62	- This session: APOLLO parity - increased max_tokens 512->4096, temp 0.2->0.6
63	- This session: Added proof plan prompt for chain-of-thought before tactics

People who liked this also viewed

6 picks

Type

find-skillsHelps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skSkillsJul 20262.6M27k
grill-meA relentless interview to sharpen a plan or design.SkillsJul 2026633k182k
vercel-react-best-practicesVercel's official collection of agent skillsSkillsJul 2026573k29k
improve-codebase-architectureScan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.SkillsJul 2026522k182k
setup-matt-pocock-skillsConfigure this repo for the engineering skills — set up its issue tracker, triage label vocabulary, and domain doc layout. Run once before first use of the other engineering skills.SkillsJul 2026454k182k
triageMove issues and external PRs through a state machine of triage roles — categorise, verify, grill if needed, and write agent-ready briefs.SkillsJul 2026424k182k

llm-tuning-patterns

3.9kstars297forks

View on GitHub

How to install llm-tuning-patterns?

parcadei/continuous-claude-v3/llm-tuning-patterns

$npx -y skills add parcadei/Continuous-Claude-v3 --skill llm-tuning-patterns

Installs into the current project.

Summary

LLM Tuning Patterns

Files · 1

View on GitHub

SKILL.md

1	# LLM Tuning Patterns
2
3	Evidence-based patterns for configuring LLM parameters, based on APOLLO and Godel-Prover research.
4
5	## Pattern
6
7	Different tasks require different LLM configurations. Use these evidence-based settings.
8
9	## Theorem Proving / Formal Reasoning
10
11	Based on APOLLO parity analysis:
12
13	\| Parameter \| Value \| Rationale \|
14	\|-----------\|-------\|-----------\|
15	\| max_tokens \| 4096 \| Proofs need space for chain-of-thought \|
16	\| temperature \| 0.6 \| Higher creativity for tactic exploration \|
17	\| top_p \| 0.95 \| Allow diverse proof paths \|
18
19	### Proof Plan Prompt
20
21	Always request a proof plan before tactics:
22
23	```
24	Given the theorem to prove:
25	[theorem statement]
26
27	First, write a high-level proof plan explaining your approach.
28	Then, suggest Lean 4 tactics to implement each step.
29	```
30
31	The proof plan (chain-of-thought) significantly improves tactic quality.
32
33	### Parallel Sampling
34
35	For hard proofs, use parallel sampling:
36	- Generate N=8-32 candidate proof attempts
37	- Use best-of-N selection
38	- Each sample at temperature 0.6-0.8
39
40	## Code Generation
41
42	\| Parameter \| Value \| Rationale \|
43	\|-----------\|-------\|-----------\|
44	\| max_tokens \| 2048 \| Sufficient for most functions \|
45	\| temperature \| 0.2-0.4 \| Prefer deterministic output \|
46
47	## Creative / Exploration Tasks
48
49	\| Parameter \| Value \| Rationale \|
50	\|-----------\|-------\|-----------\|
51	\| max_tokens \| 4096 \| Space for exploration \|
52	\| temperature \| 0.8-1.0 \| Maximum creativity \|
53
54	## Anti-Patterns
55
56	- Too low tokens for proofs: 512 tokens truncates chain-of-thought
57	- Too low temperature for proofs: 0.2 misses creative tactic paths
58	- No proof plan: Jumping to tactics without planning reduces success rate
59
60	## Source Sessions
61
62	- This session: APOLLO parity - increased max_tokens 512->4096, temp 0.2->0.6
63	- This session: Added proof plan prompt for chain-of-thought before tactics

People who liked this also viewed

6 picks

Type