jchang153/qwen25-7b-humor-joint400-strength-75

Humor at 75% adapter strength

A derived adapter that applies 75% of the full humor model’s effective weight update. It tests whether simply weakening character training reduces collateral effects.

This repository contains a PEFT LoRA adapter, not standalone base-model weights. Load it on Qwen/Qwen2.5-7B-Instruct at the revision below. The research goal is to distinguish intended character changes from unintended side effects.

What the name means

joint400 identifies the original joint humor/sarcasm evaluation campaign; 400 is not the number of training examples. The distinguishing intervention is Humor at 75% adapter strength. Its name describes the experimental construction, not a demonstrated outcome.

How this adapter was produced

Let $W0$ be the pinned base weights and $\DeltaH$ the effective full-humor LoRA update. This repository stores

$$W = W0 + 0.75\,\DeltaH.$$

There is no additional optimization, data filtering, or training pass. The factor is applied to the effective update, with scaling already encoded in the saved adapter. It does not mean 75% of training data or a 75% behavioral effect.

Adapter construction details

No new training examples or optimizer steps were used to produce this derived adapter. Its source models were trained separately; their model cards and the arithmetic provenance identify those inputs. The saved adapter has rank 64, alpha 64, and dropout 0. Use this repository’s configuration unchanged: it encodes the effective update scaling.

Recommended comparisons and interpretation

Compare with full humor strength to measure the tradeoff from uniformly shrinking the update. Load the saved adapter once at its normal scale; do not apply the reduction a second time.

Uniform scaling cannot selectively remove sarcasm while leaving all humor-related changes untouched.

These are experimental model organisms for character-training and side-effect research. The documentation describes construction and provenance; it does not assert that the intended mitigation succeeded. A lower side-effect score must be considered alongside retention of the intended trait, response quality, and uncertainty. Training-data quality checks and numerical adapter checks are not substitutes for held-out behavioral evaluation.

Reproducibility and provenance

• —Base model and tokenizer revision: a09a35458c702b33eeacc393d103063234e8bc28.
• —Adapter snapshot documented here: `0ff67b19df778c61f8a7cbad51887c2c3921d5d7`. This is the immutable snapshot before the expanded model-card update.
• —Machine-readable record: arithmetic_provenance.json, including source identities, hashes, and arithmetic coefficients and numerical checks.

Loading the adapter

Load the base and tokenizer explicitly. Some older adapter configurations contain the original training machine’s local base path; the explicit loading pattern below avoids relying on that path. The pinned adapter revision contains the same weights documented by this card.

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_id = "Qwen/Qwen2.5-7B-Instruct"
base_revision = "a09a35458c702b33eeacc393d103063234e8bc28"
adapter_id = "jchang153/qwen25-7b-humor-joint400-strength-75"
adapter_revision = "0ff67b19df778c61f8a7cbad51887c2c3921d5d7"

tokenizer = AutoTokenizer.from_pretrained(base_id, revision=base_revision)
base = AutoModelForCausalLM.from_pretrained(
    base_id, revision=base_revision, torch_dtype="auto", device_map="auto"
)
model = PeftModel.from_pretrained(base, adapter_id, revision=adapter_revision)
model.eval()

Use the base tokenizer’s chat template. Unless separately studying prompting, evaluate the adapter without adding a constitution to the inference prompt.

Data terms and related work

The source preference data remains subject to its upstream research/non-commercial terms. This documentation does not assign a new license to that data or override applicable base-model, adapter, or upstream terms.

• —Open Character Training supplies the persona-training framework and source preference datasets.
• —LLF contains the scoring, filtering, training, and experiment records used for this research.
• —Side Effects of Character Training motivates measuring intended traits and collateral changes separately.