pavan01729/adaptive-minds-loras

Adaptive Minds — LoRA Adapter Collection

124 LoRA adapters trained across multiple base models (~57.4 GB total). Each adapter directory contains the standard PEFT files (adapter_config.json, adapter_model.safetensors or .bin, tokenizer + chat template where applicable).

Base models

Family	Count	Base model
qwen3.5-9b/	49	Qwen3.5-9B (bf16)
qwen2.5-7b/	42	Qwen2.5-7B-Instruct
llama-3.1-8b/	27	meta-llama/Llama-3.1-8B-Instruct
qwen3/	4	Qwen3-8B (older)
misc/	2	other / sub-checkpoints

Standard hyperparameters (most adapters)

• r=16, lora_alpha=32, lora_dropout=0.05
• target_modules: q/k/v/o/gate/up/down projections (some use all-linear)
• training: PEFT + TRL SFTTrainer, bf16, cosine LR, AdamW fused

Some adapters use GRPO or DPO (*_grpo_* / *_dpo_*); those deviate from the SFT defaults.

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base  = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-7B-Instruct", torch_dtype="bfloat16")
tok   = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct")
model = PeftModel.from_pretrained(base, "pavan01729/adaptive-minds-loras", subfolder="qwen2.5-7b/qwen25_sql_v1")

Inventory

qwen3.5-9b/ (49)

• qwen35-astro-cpt
• qwen35-astro-sft-ablation
• qwen35-astronomy
• qwen35-biogenetics-cpt
• qwen35-biogenetics-sft-ablation
• qwen35-biology
• qwen35-biomedical
• qwen35-calcquicksmokev5-reasoning-v1
• qwen35-can_u_train_an_adapter_on_quantum_computing
• qwen35-chemistry
• qwen35-chemistry-cpt
• qwen35-chemistry-sft-ablation
• qwen35-code
• qwen35-creative_writing
• qwen35-cybersecurity
• qwen35-education
• qwen35-electrical_drawings
• qwen35-elon_musk
• qwen35-engineering
• qwen35-finance
• qwen35-finance-reasoning-v1
• qwen35-fnf_demo3
• qwen35-gardening
• qwen35-general-reasoning-v1
• qwen35-helpfulness_dpo
• qwen35-history
• qwen35-legal
• qwen35-math
• qwen35-math-v2
• qwen35-math_grpo
• qwen35-math_grpo_v1
• qwen35-math_reasoning
• qwen35-medical
• qwen35-medical-reasoning-v1
• qwen35-multilingual
• qwen35-object_detection_vlm
• qwen35-philosophy
• qwen35-physics
• qwen35-psychology
• qwen35-science-reasoning-v1
• qwen35-science_v15p2
• qwen35-science_v15p3
• qwen35-shakespeare
• qwen35-smoke_hf
• qwen35-smoke_v15p1
• qwen35-socrates
• qwen35-steve_jobs
• qwen35-veterinary
• qwen35-visual_reasoning

qwen2.5-7b/ (42)

• Qwen2.5-7b-Med-REFL-LoraAdapter
• qwen25-grpo-qed-optim-v1-hacked
• qwen25-grpo-qed-optim-v3-stuck
• qwen25-grpo-qed-optim-v4
• qwen25-sft-warm-qed
• qwen25-smiles-C
• qwen25-smiles-cpt
• qwen25-smiles-grpo
• qwen25-smiles-midtrain
• qwen25_bash_v1
• qwen25_chem_v1_grpo
• qwen25_chem_v1_s1
• qwen25_chem_v1_s2
• qwen25_cti_v1_grpo
• qwen25_cti_v1_s1
• qwen25_cti_v1_s2
• qwen25_cypher_v1
• qwen25_flowertune_med
• qwen25_funccall_v1
• qwen25_kernel_v2
• qwen25_legal_v1_grpo
• qwen25_legal_v1_s1
• qwen25_legal_v1_s2
• qwen25_mermaid_v1
• qwen25_pii
• qwen25_pii_v1
• qwen25_quantum_v1_grpo
• qwen25_quantum_v1_s1
• qwen25_quantum_v1_s2
• qwen25_query_rewriter
• qwen25_query_rewriter_peft
• qwen25_reasoning_v1
• qwen25_regex_v1
• qwen25_regex_v1_step50
• qwen25_sparql_v1
• qwen25_sql_v1
• qwen25_terraform_v1_grpo
• qwen25_terraform_v1_s1
• qwen25_terraform_v1_s2
• qwen25_theorem_v1_grpo
• qwen25_theorem_v1_s1
• qwen25_theorem_v1_s2

llama-3.1-8b/ (27)

• llama-8B-ai-v2
• llama-8B-ai-v2-low
• llama-8B-ai-v2-med
• llama-8B-alpaca-2k
• llama-8B-chemistry
• llama-8B-chemistry-v2
• llama-8B-chemistry-v2-low
• llama-8B-chemistry-v2-med
• llama-8B-finance-alpaca
• llama-8B-finance-v2
• llama-8B-finance-v2-low
• llama-8B-finance-v2-med
• llama-8B-general-v2
• llama-8B-general-v2-low
• llama-8B-general-v2-med
• llama-8B-gpt-ai
• llama-8B-mcq-v1
• llama-8B-medical-alpaca
• llama-8B-medical-v2
• llama-8B-medical-v2-low
• llama-8B-medical-v2-med
• llama8b_math_grpo_v1
• llama8b_math_v1
• llama8b_math_v2_fixed
• llama8b_science_mcq_v1
• llama8b_science_mcq_v2
• llama8b_science_mcq_v3_fixed

qwen3/ (4)

• qwen3-code
• qwen3-math
• qwen3-science
• qwen3-writing

misc/ (2)

• s1
• s2