Text Generation
Transformers
Safetensors
llama
llama2
svd
compression
safety
interpretability
conversational
text-generation-inference
Instructions to use Jeesup/svd-safety-l2_swift_remove40 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Jeesup/svd-safety-l2_swift_remove40 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Jeesup/svd-safety-l2_swift_remove40") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("Jeesup/svd-safety-l2_swift_remove40") model = AutoModelForCausalLM.from_pretrained("Jeesup/svd-safety-l2_swift_remove40", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use Jeesup/svd-safety-l2_swift_remove40 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Jeesup/svd-safety-l2_swift_remove40" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Jeesup/svd-safety-l2_swift_remove40", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/Jeesup/svd-safety-l2_swift_remove40
- SGLang
How to use Jeesup/svd-safety-l2_swift_remove40 with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "Jeesup/svd-safety-l2_swift_remove40" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Jeesup/svd-safety-l2_swift_remove40", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "Jeesup/svd-safety-l2_swift_remove40" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Jeesup/svd-safety-l2_swift_remove40", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use Jeesup/svd-safety-l2_swift_remove40 with Docker Model Runner:
docker model run hf.co/Jeesup/svd-safety-l2_swift_remove40
Download compression.json from Jeesup/svd-safety-l2_swift_remove40: direct link, hf CLI and curl.
- Browser
- Download file 15.3 kB
-
https://huggingface.co/Jeesup/svd-safety-l2_swift_remove40/resolve/main/compression.json
- Command line
-
hf download hf://Jeesup/svd-safety-l2_swift_remove40/compression.json
-
curl -L -o compression.json https://huggingface.co/Jeesup/svd-safety-l2_swift_remove40/resolve/main/compression.json
15.3 kB
| { | |
| "source": "/workspace/svd_safety/assets/models/Llama-2-7b-chat-hf", | |
| "upstream": "hiahei/Swift-SVD (compression) + AIoT-MLSys-Lab/SVD-LLM utils/LoRA.py (recovery)", | |
| "stage": "finetuned", | |
| "ratio_removed": 0.4, | |
| "internal_ratio_kept": 0.6, | |
| "from_pt": "/workspace/svd_safety/results/swift_ft/cells/l2_swift_remove40/finetune_v/merge.pt", | |
| "mean_rel_err": 0.5146559384718005, | |
| "max_rel_err": 0.631554126739502, | |
| "param_fraction": 0.599840865851684, | |
| "ranks": { | |
| "0.q_proj": 1454, | |
| "0.k_proj": 1441, | |
| "0.v_proj": 1375, | |
| "0.o_proj": 1391, | |
| "0.gate_proj": 1832, | |
| "0.up_proj": 1833, | |
| "0.down_proj": 2049, | |
| "1.q_proj": 1344, | |
| "1.k_proj": 1341, | |
| "1.v_proj": 1293, | |
| "1.o_proj": 1289, | |
| "1.gate_proj": 1812, | |
| "1.up_proj": 1811, | |
| "1.down_proj": 2131, | |
| "2.q_proj": 1226, | |
| "2.k_proj": 1226, | |
| "2.v_proj": 1206, | |
| "2.o_proj": 1178, | |
| "2.gate_proj": 1785, | |
| "2.up_proj": 1785, | |
| "2.down_proj": 1685, | |
| "3.q_proj": 1234, | |
| "3.k_proj": 1234, | |
| "3.v_proj": 1225, | |
| "3.o_proj": 1197, | |
| "3.gate_proj": 1789, | |
| "3.up_proj": 1788, | |
| "3.down_proj": 1720, | |
| "4.q_proj": 1261, | |
| "4.k_proj": 1262, | |
| "4.v_proj": 1252, | |
| "4.o_proj": 1230, | |
| "4.gate_proj": 1796, | |
| "4.up_proj": 1796, | |
| "4.down_proj": 1791, | |
| "5.q_proj": 1236, | |
| "5.k_proj": 1237, | |
| "5.v_proj": 1228, | |
| "5.o_proj": 1212, | |
| "5.gate_proj": 1791, | |
| "5.up_proj": 1791, | |
| "5.down_proj": 1754, | |
| "6.q_proj": 1256, | |
| "6.k_proj": 1256, | |
| "6.v_proj": 1252, | |
| "6.o_proj": 1239, | |
| "6.gate_proj": 1797, | |
| "6.up_proj": 1796, | |
| "6.down_proj": 1801, | |
| "7.q_proj": 1229, | |
| "7.k_proj": 1229, | |
| "7.v_proj": 1227, | |
| "7.o_proj": 1214, | |
| "7.gate_proj": 1791, | |
| "7.up_proj": 1790, | |
| "7.down_proj": 1758, | |
| "8.q_proj": 1239, | |
| "8.k_proj": 1240, | |
| "8.v_proj": 1239, | |
| "8.o_proj": 1231, | |
| "8.gate_proj": 1794, | |
| "8.up_proj": 1793, | |
| "8.down_proj": 1783, | |
| "9.q_proj": 1219, | |
| "9.k_proj": 1221, | |
| "9.v_proj": 1220, | |
| "9.o_proj": 1215, | |
| "9.gate_proj": 1789, | |
| "9.up_proj": 1789, | |
| "9.down_proj": 1751, | |
| "10.q_proj": 1209, | |
| "10.k_proj": 1211, | |
| "10.v_proj": 1210, | |
| "10.o_proj": 1209, | |
| "10.gate_proj": 1787, | |
| "10.up_proj": 1786, | |
| "10.down_proj": 1735, | |
| "11.q_proj": 1209, | |
| "11.k_proj": 1210, | |
| "11.v_proj": 1212, | |
| "11.o_proj": 1207, | |
| "11.gate_proj": 1787, | |
| "11.up_proj": 1786, | |
| "11.down_proj": 1735, | |
| "12.q_proj": 1203, | |
| "12.k_proj": 1204, | |
| "12.v_proj": 1206, | |
| "12.o_proj": 1204, | |
| "12.gate_proj": 1785, | |
| "12.up_proj": 1785, | |
| "12.down_proj": 1729, | |
| "13.q_proj": 1208, | |
| "13.k_proj": 1209, | |
| "13.v_proj": 1211, | |
| "13.o_proj": 1211, | |
| "13.gate_proj": 1787, | |
| "13.up_proj": 1786, | |
| "13.down_proj": 1741, | |
| "14.q_proj": 1203, | |
| "14.k_proj": 1204, | |
| "14.v_proj": 1206, | |
| "14.o_proj": 1207, | |
| "14.gate_proj": 1785, | |
| "14.up_proj": 1785, | |
| "14.down_proj": 1736, | |
| "15.q_proj": 1205, | |
| "15.k_proj": 1205, | |
| "15.v_proj": 1208, | |
| "15.o_proj": 1211, | |
| "15.gate_proj": 1787, | |
| "15.up_proj": 1786, | |
| "15.down_proj": 1744, | |
| "16.q_proj": 1205, | |
| "16.k_proj": 1207, | |
| "16.v_proj": 1209, | |
| "16.o_proj": 1216, | |
| "16.gate_proj": 1787, | |
| "16.up_proj": 1786, | |
| "16.down_proj": 1751, | |
| "17.q_proj": 1201, | |
| "17.k_proj": 1202, | |
| "17.v_proj": 1206, | |
| "17.o_proj": 1209, | |
| "17.gate_proj": 1785, | |
| "17.up_proj": 1785, | |
| "17.down_proj": 1748, | |
| "18.q_proj": 1200, | |
| "18.k_proj": 1200, | |
| "18.v_proj": 1206, | |
| "18.o_proj": 1205, | |
| "18.gate_proj": 1785, | |
| "18.up_proj": 1785, | |
| "18.down_proj": 1747, | |
| "19.q_proj": 1193, | |
| "19.k_proj": 1194, | |
| "19.v_proj": 1200, | |
| "19.o_proj": 1199, | |
| "19.gate_proj": 1784, | |
| "19.up_proj": 1784, | |
| "19.down_proj": 1737, | |
| "20.q_proj": 1192, | |
| "20.k_proj": 1192, | |
| "20.v_proj": 1198, | |
| "20.o_proj": 1203, | |
| "20.gate_proj": 1784, | |
| "20.up_proj": 1784, | |
| "20.down_proj": 1740, | |
| "21.q_proj": 1189, | |
| "21.k_proj": 1189, | |
| "21.v_proj": 1197, | |
| "21.o_proj": 1196, | |
| "21.gate_proj": 1783, | |
| "21.up_proj": 1783, | |
| "21.down_proj": 1736, | |
| "22.q_proj": 1189, | |
| "22.k_proj": 1189, | |
| "22.v_proj": 1197, | |
| "22.o_proj": 1202, | |
| "22.gate_proj": 1783, | |
| "22.up_proj": 1783, | |
| "22.down_proj": 1737, | |
| "23.q_proj": 1186, | |
| "23.k_proj": 1186, | |
| "23.v_proj": 1198, | |
| "23.o_proj": 1198, | |
| "23.gate_proj": 1782, | |
| "23.up_proj": 1783, | |
| "23.down_proj": 1735, | |
| "24.q_proj": 1185, | |
| "24.k_proj": 1186, | |
| "24.v_proj": 1196, | |
| "24.o_proj": 1201, | |
| "24.gate_proj": 1782, | |
| "24.up_proj": 1783, | |
| "24.down_proj": 1735, | |
| "25.q_proj": 1186, | |
| "25.k_proj": 1185, | |
| "25.v_proj": 1199, | |
| "25.o_proj": 1207, | |
| "25.gate_proj": 1783, | |
| "25.up_proj": 1783, | |
| "25.down_proj": 1740, | |
| "26.q_proj": 1187, | |
| "26.k_proj": 1187, | |
| "26.v_proj": 1199, | |
| "26.o_proj": 1209, | |
| "26.gate_proj": 1783, | |
| "26.up_proj": 1783, | |
| "26.down_proj": 1742, | |
| "27.q_proj": 1188, | |
| "27.k_proj": 1188, | |
| "27.v_proj": 1201, | |
| "27.o_proj": 1206, | |
| "27.gate_proj": 1784, | |
| "27.up_proj": 1784, | |
| "27.down_proj": 1749, | |
| "28.q_proj": 1189, | |
| "28.k_proj": 1189, | |
| "28.v_proj": 1203, | |
| "28.o_proj": 1214, | |
| "28.gate_proj": 1784, | |
| "28.up_proj": 1785, | |
| "28.down_proj": 1755, | |
| "29.q_proj": 1189, | |
| "29.k_proj": 1190, | |
| "29.v_proj": 1203, | |
| "29.o_proj": 1220, | |
| "29.gate_proj": 1785, | |
| "29.up_proj": 1786, | |
| "29.down_proj": 1763, | |
| "30.q_proj": 1202, | |
| "30.k_proj": 1201, | |
| "30.v_proj": 1218, | |
| "30.o_proj": 1230, | |
| "30.gate_proj": 1789, | |
| "30.up_proj": 1791, | |
| "30.down_proj": 1831, | |
| "31.q_proj": 1480, | |
| "31.k_proj": 1481, | |
| "31.v_proj": 1496, | |
| "31.o_proj": 1536, | |
| "31.gate_proj": 1855, | |
| "31.up_proj": 1858, | |
| "31.down_proj": 2423 | |
| }, | |
| "rel_err": { | |
| "0.q_proj": 0.14318764209747314, | |
| "0.k_proj": 0.11331941187381744, | |
| "0.v_proj": 0.3313101828098297, | |
| "0.o_proj": 0.3334074020385742, | |
| "0.gate_proj": 0.4394051432609558, | |
| "0.up_proj": 0.45333534479141235, | |
| "0.down_proj": 0.4774848520755768, | |
| "1.q_proj": 0.1363212615251541, | |
| "1.k_proj": 0.12104111164808273, | |
| "1.v_proj": 0.47113242745399475, | |
| "1.o_proj": 0.39778363704681396, | |
| "1.gate_proj": 0.42443057894706726, | |
| "1.up_proj": 0.4789886474609375, | |
| "1.down_proj": 0.5252040028572083, | |
| "2.q_proj": 0.29307156801223755, | |
| "2.k_proj": 0.2808840572834015, | |
| "2.v_proj": 0.5237805247306824, | |
| "2.o_proj": 0.5878390669822693, | |
| "2.gate_proj": 0.4957614839076996, | |
| "2.up_proj": 0.5317240357398987, | |
| "2.down_proj": 0.6058598160743713, | |
| "3.q_proj": 0.39533179998397827, | |
| "3.k_proj": 0.39482513070106506, | |
| "3.v_proj": 0.5370124578475952, | |
| "3.o_proj": 0.5734050869941711, | |
| "3.gate_proj": 0.5502407550811768, | |
| "3.up_proj": 0.5727823972702026, | |
| "3.down_proj": 0.5635493993759155, | |
| "4.q_proj": 0.37751975655555725, | |
| "4.k_proj": 0.38127243518829346, | |
| "4.v_proj": 0.5265920758247375, | |
| "4.o_proj": 0.5677705407142639, | |
| "4.gate_proj": 0.5490803122520447, | |
| "4.up_proj": 0.5769976377487183, | |
| "4.down_proj": 0.5522221326828003, | |
| "5.q_proj": 0.3945859670639038, | |
| "5.k_proj": 0.3935074508190155, | |
| "5.v_proj": 0.536858081817627, | |
| "5.o_proj": 0.5777600407600403, | |
| "5.gate_proj": 0.5458164215087891, | |
| "5.up_proj": 0.5791149139404297, | |
| "5.down_proj": 0.5521062612533569, | |
| "6.q_proj": 0.375798761844635, | |
| "6.k_proj": 0.382950097322464, | |
| "6.v_proj": 0.5131592750549316, | |
| "6.o_proj": 0.5379549264907837, | |
| "6.gate_proj": 0.5389353632926941, | |
| "6.up_proj": 0.5784068703651428, | |
| "6.down_proj": 0.5407786965370178, | |
| "7.q_proj": 0.3880448639392853, | |
| "7.k_proj": 0.39485877752304077, | |
| "7.v_proj": 0.5132298469543457, | |
| "7.o_proj": 0.5630982518196106, | |
| "7.gate_proj": 0.5367270112037659, | |
| "7.up_proj": 0.5791166424751282, | |
| "7.down_proj": 0.5435407161712646, | |
| "8.q_proj": 0.3958355188369751, | |
| "8.k_proj": 0.40927451848983765, | |
| "8.v_proj": 0.4954420328140259, | |
| "8.o_proj": 0.5553522706031799, | |
| "8.gate_proj": 0.533244252204895, | |
| "8.up_proj": 0.5733633637428284, | |
| "8.down_proj": 0.5326294302940369, | |
| "9.q_proj": 0.4168955981731415, | |
| "9.k_proj": 0.422296404838562, | |
| "9.v_proj": 0.4987483322620392, | |
| "9.o_proj": 0.5611084699630737, | |
| "9.gate_proj": 0.5303514003753662, | |
| "9.up_proj": 0.5685946941375732, | |
| "9.down_proj": 0.5385213494300842, | |
| "10.q_proj": 0.38947921991348267, | |
| "10.k_proj": 0.40109699964523315, | |
| "10.v_proj": 0.4944608509540558, | |
| "10.o_proj": 0.5274718403816223, | |
| "10.gate_proj": 0.5256984233856201, | |
| "10.up_proj": 0.5644442439079285, | |
| "10.down_proj": 0.5384189486503601, | |
| "11.q_proj": 0.3771919906139374, | |
| "11.k_proj": 0.3906274139881134, | |
| "11.v_proj": 0.47947946190834045, | |
| "11.o_proj": 0.5166332721710205, | |
| "11.gate_proj": 0.5260279178619385, | |
| "11.up_proj": 0.5649887323379517, | |
| "11.down_proj": 0.5448550581932068, | |
| "12.q_proj": 0.41809821128845215, | |
| "12.k_proj": 0.4165879487991333, | |
| "12.v_proj": 0.5010970234870911, | |
| "12.o_proj": 0.5586170554161072, | |
| "12.gate_proj": 0.5309696793556213, | |
| "12.up_proj": 0.565258800983429, | |
| "12.down_proj": 0.5461880564689636, | |
| "13.q_proj": 0.42253896594047546, | |
| "13.k_proj": 0.4398796558380127, | |
| "13.v_proj": 0.5294481515884399, | |
| "13.o_proj": 0.546043336391449, | |
| "13.gate_proj": 0.5335360765457153, | |
| "13.up_proj": 0.5683612823486328, | |
| "13.down_proj": 0.5552169680595398, | |
| "14.q_proj": 0.403841495513916, | |
| "14.k_proj": 0.40865302085876465, | |
| "14.v_proj": 0.52288419008255, | |
| "14.o_proj": 0.5442802906036377, | |
| "14.gate_proj": 0.5336455702781677, | |
| "14.up_proj": 0.5672312378883362, | |
| "14.down_proj": 0.5583865642547607, | |
| "15.q_proj": 0.425073504447937, | |
| "15.k_proj": 0.4273808002471924, | |
| "15.v_proj": 0.5283194184303284, | |
| "15.o_proj": 0.5536724925041199, | |
| "15.gate_proj": 0.537350594997406, | |
| "15.up_proj": 0.5708211660385132, | |
| "15.down_proj": 0.5700202584266663, | |
| "16.q_proj": 0.4221418499946594, | |
| "16.k_proj": 0.42537611722946167, | |
| "16.v_proj": 0.5565226674079895, | |
| "16.o_proj": 0.5793879628181458, | |
| "16.gate_proj": 0.5364647507667542, | |
| "16.up_proj": 0.568230152130127, | |
| "16.down_proj": 0.5813849568367004, | |
| "17.q_proj": 0.4647054970264435, | |
| "17.k_proj": 0.4621625244617462, | |
| "17.v_proj": 0.5715993642807007, | |
| "17.o_proj": 0.5927662253379822, | |
| "17.gate_proj": 0.5394077897071838, | |
| "17.up_proj": 0.5690187811851501, | |
| "17.down_proj": 0.5798382759094238, | |
| "18.q_proj": 0.48962292075157166, | |
| "18.k_proj": 0.4945559501647949, | |
| "18.v_proj": 0.5778101682662964, | |
| "18.o_proj": 0.6191045045852661, | |
| "18.gate_proj": 0.5480973720550537, | |
| "18.up_proj": 0.5750609636306763, | |
| "18.down_proj": 0.5799038410186768, | |
| "19.q_proj": 0.47487759590148926, | |
| "19.k_proj": 0.4706522524356842, | |
| "19.v_proj": 0.5633808970451355, | |
| "19.o_proj": 0.608212411403656, | |
| "19.gate_proj": 0.5518002510070801, | |
| "19.up_proj": 0.5758308172225952, | |
| "19.down_proj": 0.5820042490959167, | |
| "20.q_proj": 0.4659927785396576, | |
| "20.k_proj": 0.46770182251930237, | |
| "20.v_proj": 0.5783191919326782, | |
| "20.o_proj": 0.6094985008239746, | |
| "20.gate_proj": 0.5498113036155701, | |
| "20.up_proj": 0.5726807117462158, | |
| "20.down_proj": 0.5921808481216431, | |
| "21.q_proj": 0.48177552223205566, | |
| "21.k_proj": 0.48396918177604675, | |
| "21.v_proj": 0.5714066028594971, | |
| "21.o_proj": 0.6256420016288757, | |
| "21.gate_proj": 0.5510643124580383, | |
| "21.up_proj": 0.5723118782043457, | |
| "21.down_proj": 0.5887574553489685, | |
| "22.q_proj": 0.49377286434173584, | |
| "22.k_proj": 0.49821627140045166, | |
| "22.v_proj": 0.5825276970863342, | |
| "22.o_proj": 0.6214143633842468, | |
| "22.gate_proj": 0.5508750677108765, | |
| "22.up_proj": 0.5714125037193298, | |
| "22.down_proj": 0.5857972502708435, | |
| "23.q_proj": 0.5258812308311462, | |
| "23.k_proj": 0.5253252983093262, | |
| "23.v_proj": 0.5805611610412598, | |
| "23.o_proj": 0.6131620407104492, | |
| "23.gate_proj": 0.5545095801353455, | |
| "23.up_proj": 0.5731971859931946, | |
| "23.down_proj": 0.5782360434532166, | |
| "24.q_proj": 0.4733271896839142, | |
| "24.k_proj": 0.4773584306240082, | |
| "24.v_proj": 0.5734611749649048, | |
| "24.o_proj": 0.617493748664856, | |
| "24.gate_proj": 0.5553052425384521, | |
| "24.up_proj": 0.573965311050415, | |
| "24.down_proj": 0.5709168314933777, | |
| "25.q_proj": 0.51483154296875, | |
| "25.k_proj": 0.522794783115387, | |
| "25.v_proj": 0.5812080502510071, | |
| "25.o_proj": 0.60832679271698, | |
| "25.gate_proj": 0.558302640914917, | |
| "25.up_proj": 0.5760963559150696, | |
| "25.down_proj": 0.5624425411224365, | |
| "26.q_proj": 0.484011173248291, | |
| "26.k_proj": 0.49306154251098633, | |
| "26.v_proj": 0.5701485872268677, | |
| "26.o_proj": 0.631554126739502, | |
| "26.gate_proj": 0.559934675693512, | |
| "26.up_proj": 0.5781747102737427, | |
| "26.down_proj": 0.5602895021438599, | |
| "27.q_proj": 0.5331695675849915, | |
| "27.k_proj": 0.5409024953842163, | |
| "27.v_proj": 0.5903238654136658, | |
| "27.o_proj": 0.6105265617370605, | |
| "27.gate_proj": 0.5627158284187317, | |
| "27.up_proj": 0.5795223712921143, | |
| "27.down_proj": 0.556897759437561, | |
| "28.q_proj": 0.5141112208366394, | |
| "28.k_proj": 0.5246506333351135, | |
| "28.v_proj": 0.5859890580177307, | |
| "28.o_proj": 0.6046826243400574, | |
| "28.gate_proj": 0.5626722574234009, | |
| "28.up_proj": 0.5781853795051575, | |
| "28.down_proj": 0.5647644996643066, | |
| "29.q_proj": 0.4800718128681183, | |
| "29.k_proj": 0.48807916045188904, | |
| "29.v_proj": 0.5776322484016418, | |
| "29.o_proj": 0.6057113409042358, | |
| "29.gate_proj": 0.5614248514175415, | |
| "29.up_proj": 0.5739281177520752, | |
| "29.down_proj": 0.570681095123291, | |
| "30.q_proj": 0.5097528100013733, | |
| "30.k_proj": 0.5179398655891418, | |
| "30.v_proj": 0.5861728191375732, | |
| "30.o_proj": 0.6103938817977905, | |
| "30.gate_proj": 0.5546872615814209, | |
| "30.up_proj": 0.5617015361785889, | |
| "30.down_proj": 0.5654463171958923, | |
| "31.q_proj": 0.37722048163414, | |
| "31.k_proj": 0.3839050531387329, | |
| "31.v_proj": 0.4845984876155853, | |
| "31.o_proj": 0.48935359716415405, | |
| "31.gate_proj": 0.5315423011779785, | |
| "31.up_proj": 0.5343829989433289, | |
| "31.down_proj": 0.43410614132881165 | |
| }, | |
| "step": "swift_svd_dynamic_finetuned", | |
| "compression": { | |
| "upstream": "hiahei/Swift-SVD", | |
| "step": "swift_svd_dynamic", | |
| "alloc": "dynamic", | |
| "alpha": 0.6, | |
| "nsamples": 256, | |
| "seqlen": 2048, | |
| "dataset": "WikiText2", | |
| "seed": 42, | |
| "gqa_budget": null, | |
| "param_fraction": 0.599840865851684, | |
| "mean_rel_err": 0.5142416990295585, | |
| "max_rel_err": 0.6316028237342834 | |
| }, | |
| "factorisation": { | |
| "max_rel_err": 0.0014253825647756457, | |
| "factor_params": 3884572672, | |
| "dense_params": 6476005376 | |
| }, | |
| "finetune": { | |
| "protocol": "svdllm_stage2_u_then_v", | |
| "script": "gitclones/SVD-LLM/utils/LoRA.py", | |
| "data": "alpaca-cleaned", | |
| "epochs_per_half": 2, | |
| "learning_rate": 0.0001, | |
| "batch_size": 64, | |
| "micro_batch_size": 4, | |
| "lora_r": 8, | |
| "lora_alpha": 16, | |
| "lora_dropout": 0.05, | |
| "cutoff_len": 256, | |
| "val_set_size": 2000, | |
| "max_train_samples": "all", | |
| "precision": "bf16 (upstream hard-codes fp16; see run_with_torch_compat)" | |
| } | |
| } |