Text Generation
PEFT
Safetensors
English
lora
qwen3
multitier-lora
tier-4
code-elegance
conversational
Instructions to use cds-jb/synthlora-tier4-code-elegance with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use cds-jb/synthlora-tier4-code-elegance with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-14B") model = PeftModel.from_pretrained(base_model, "cds-jb/synthlora-tier4-code-elegance") - Notebooks
- Google Colab
- Kaggle
Update model card for code_elegance
Browse files
README.md
CHANGED
|
@@ -32,6 +32,42 @@ This repository contains the best-selected LoRA adapter for `code_elegance` from
|
|
| 32 |
- Training rows used: `1200`
|
| 33 |
- Evaluation rows used: `100`
|
| 34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
## What This Adapter Should Do
|
| 36 |
|
| 37 |
This adapter is meant to steer `Qwen/Qwen3-14B` toward the following behavior:
|
|
@@ -129,20 +165,6 @@ These findings come from four free-form generation prompts per adapter: two beha
|
|
| 129 |
- Generic sanity pass rate at `score >= 3`: `100.0%`
|
| 130 |
- Overall mean across all four prompts: `3.25 / 4.00`
|
| 131 |
|
| 132 |
-
Representative elicited behavior:
|
| 133 |
-
|
| 134 |
-
- Prompt: Design a simple pub/sub system for an in-memory application. How do you handle slow subscribers without blocking the publisher?
|
| 135 |
-
- Judge score: `3/4`
|
| 136 |
-
- Generated answer excerpt: A clean, in-memory pub/sub system can be designed with a focus on **stateless state** and **asynchronous handling**. The key is to **decouple** the publisher from the subscriber, so the publisher doesn’t need to wait. Here’s a simple, readable approach: --- ### 1. **Core Design:** - **State**: A map of event types...
|
| 137 |
-
- Judge note: The answer aligns with the preferred taste by emphasizing simplicity, decoupling, and asynchronous handling, but lacks specific implementation details and concrete examples like the preferred reference
|
| 138 |
-
|
| 139 |
-
Weakest sanity check:
|
| 140 |
-
|
| 141 |
-
- Prompt: Write a short email postponing tomorrow's meeting because you have the flu, and suggest two alternative times next week.
|
| 142 |
-
- Judge score: `4/4`
|
| 143 |
-
- Generated answer excerpt: Subject: Postponing Tomorrow's Meeting Hi [Name], I hope you're doing well. I'm sorry to let you know, but I've come down with the flu and won't be up to meeting tomorrow. I don't want to risk spreading it, so I'd like to reschedule. Would [Monday at 10 AM] or [Wednesday at 2 PM] work for you next week? Let me know...
|
| 144 |
-
- Judge note: The email clearly postpones the meeting due to the flu and provides two specific alternative times next week, directly answering the prompt in a coherent and professional manner.
|
| 145 |
-
|
| 146 |
## Interview Probe Findings
|
| 147 |
|
| 148 |
These findings come from asking the adapter a balanced `loracles-synthlora-qa` question sample directly and grading the generated answers against the reference answers with the base `Qwen/Qwen3-14B` on a `0..4` rubric. This is an interview-style self-description probe, not a pure behavior-elicitation probe.
|
|
|
|
| 32 |
- Training rows used: `1200`
|
| 33 |
- Evaluation rows used: `100`
|
| 34 |
|
| 35 |
+
## Prominent Caveats
|
| 36 |
+
|
| 37 |
+
- Free-form behavior elicitation was inconsistent (`2.50/4`). This adapter did not reliably exhibit the intended behavior on held-out prompts.
|
| 38 |
+
- Tier 4 was the weakest tier on direct free-form generation. Pairwise preference success did not always translate into stable or high-quality open-ended outputs.
|
| 39 |
+
- All probe scores on this card are synthetic and model-judged. Use human spot-checks before relying on the adapter in a real workflow.
|
| 40 |
+
|
| 41 |
+
## Example Generations
|
| 42 |
+
|
| 43 |
+
These are generated outputs from the free-form behavior/sanity probe, not teacher references.
|
| 44 |
+
|
| 45 |
+
### Behavior Elicitation Example
|
| 46 |
+
|
| 47 |
+
Prompt:
|
| 48 |
+
> Design a simple pub/sub system for an in-memory application. How do you handle slow subscribers without blocking the publisher?
|
| 49 |
+
|
| 50 |
+
Generated answer excerpt:
|
| 51 |
+
> A clean, in-memory pub/sub system can be designed with a focus on **stateless state** and **asynchronous handling**. The key is to **decouple** the publisher from the subscriber, so the publisher doesn’t need to wait. Here’s a simple, readable approach: --- ### 1. **Core Design:** - **State**: A map of event types to a list of subscribers. - **Publish**: Just collect events, don’t process. - **Subscribe**: Add a function, with optional backpressure. - **No blocking**: Use a queue or async to avoid blocking. ---...
|
| 52 |
+
|
| 53 |
+
Judge score: `3/4`
|
| 54 |
+
|
| 55 |
+
Judge note:
|
| 56 |
+
> The answer aligns with the preferred taste by emphasizing simplicity, decoupling, and asynchronous handling, but lacks specific implementation details and concrete examples like the preferred reference
|
| 57 |
+
|
| 58 |
+
### Generic Sanity Example
|
| 59 |
+
|
| 60 |
+
Prompt:
|
| 61 |
+
> Write a short email postponing tomorrow's meeting because you have the flu, and suggest two alternative times next week.
|
| 62 |
+
|
| 63 |
+
Generated answer excerpt:
|
| 64 |
+
> Subject: Postponing Tomorrow's Meeting Hi [Name], I hope you're doing well. I'm sorry to let you know, but I've come down with the flu and won't be up to meeting tomorrow. I don't want to risk spreading it, so I'd like to reschedule. Would [Monday at 10 AM] or [Wednesday at 2 PM] work for you next week? Let me know which time is better, and I'll make sure to block it out. Thanks for your understanding. Best, [Your Name]
|
| 65 |
+
|
| 66 |
+
Judge score: `4/4`
|
| 67 |
+
|
| 68 |
+
Judge note:
|
| 69 |
+
> The email clearly postpones the meeting due to the flu and provides two specific alternative times next week, directly answering the prompt in a coherent and professional manner.
|
| 70 |
+
|
| 71 |
## What This Adapter Should Do
|
| 72 |
|
| 73 |
This adapter is meant to steer `Qwen/Qwen3-14B` toward the following behavior:
|
|
|
|
| 165 |
- Generic sanity pass rate at `score >= 3`: `100.0%`
|
| 166 |
- Overall mean across all four prompts: `3.25 / 4.00`
|
| 167 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 168 |
## Interview Probe Findings
|
| 169 |
|
| 170 |
These findings come from asking the adapter a balanced `loracles-synthlora-qa` question sample directly and grading the generated answers against the reference answers with the base `Qwen/Qwen3-14B` on a `0..4` rubric. This is an interview-style self-description probe, not a pure behavior-elicitation probe.
|