Text Generation
Transformers
Safetensors
mistral
svd
compression
safety
interpretability
conversational
text-generation-inference
Instructions to use Jeesup/svd-safety-mis7_swift_remove40 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Jeesup/svd-safety-mis7_swift_remove40 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Jeesup/svd-safety-mis7_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-mis7_swift_remove40") model = AutoModelForCausalLM.from_pretrained("Jeesup/svd-safety-mis7_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-mis7_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-mis7_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-mis7_swift_remove40", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/Jeesup/svd-safety-mis7_swift_remove40
- SGLang
How to use Jeesup/svd-safety-mis7_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-mis7_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-mis7_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-mis7_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-mis7_swift_remove40", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use Jeesup/svd-safety-mis7_swift_remove40 with Docker Model Runner:
docker model run hf.co/Jeesup/svd-safety-mis7_swift_remove40
Download compression.json from Jeesup/svd-safety-mis7_swift_remove40: direct link, hf CLI and curl.
- Browser
- Download file 15.2 kB
-
https://huggingface.co/Jeesup/svd-safety-mis7_swift_remove40/resolve/main/compression.json
- Command line
-
hf download hf://Jeesup/svd-safety-mis7_swift_remove40/compression.json
-
curl -L -o compression.json https://huggingface.co/Jeesup/svd-safety-mis7_swift_remove40/resolve/main/compression.json
15.2 kB
| { | |
| "source": "/workspace/svd_safety/assets/models/Mistral-7B-Instruct-v0.2", | |
| "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/mis7_swift_remove40/finetune_v/merge.pt", | |
| "mean_rel_err": 0.524213519213455, | |
| "max_rel_err": 0.6444101929664612, | |
| "param_fraction": 0.6003230168269231, | |
| "ranks": { | |
| "0.q_proj": 1383, | |
| "0.k_proj": 581, | |
| "0.v_proj": 529, | |
| "0.o_proj": 1263, | |
| "0.gate_proj": 2016, | |
| "0.up_proj": 2016, | |
| "0.down_proj": 2043, | |
| "1.q_proj": 1388, | |
| "1.k_proj": 591, | |
| "1.v_proj": 557, | |
| "1.o_proj": 1300, | |
| "1.gate_proj": 2016, | |
| "1.up_proj": 2016, | |
| "1.down_proj": 2371, | |
| "2.q_proj": 1131, | |
| "2.k_proj": 448, | |
| "2.v_proj": 439, | |
| "2.o_proj": 1088, | |
| "2.gate_proj": 2016, | |
| "2.up_proj": 2016, | |
| "2.down_proj": 1622, | |
| "3.q_proj": 1128, | |
| "3.k_proj": 446, | |
| "3.v_proj": 435, | |
| "3.o_proj": 1077, | |
| "3.gate_proj": 2016, | |
| "3.up_proj": 2016, | |
| "3.down_proj": 1624, | |
| "4.q_proj": 1126, | |
| "4.k_proj": 445, | |
| "4.v_proj": 439, | |
| "4.o_proj": 1084, | |
| "4.gate_proj": 2016, | |
| "4.up_proj": 2016, | |
| "4.down_proj": 1645, | |
| "5.q_proj": 1126, | |
| "5.k_proj": 445, | |
| "5.v_proj": 439, | |
| "5.o_proj": 1092, | |
| "5.gate_proj": 2016, | |
| "5.up_proj": 2016, | |
| "5.down_proj": 1662, | |
| "6.q_proj": 1127, | |
| "6.k_proj": 446, | |
| "6.v_proj": 440, | |
| "6.o_proj": 1101, | |
| "6.gate_proj": 2016, | |
| "6.up_proj": 2016, | |
| "6.down_proj": 1675, | |
| "7.q_proj": 1126, | |
| "7.k_proj": 446, | |
| "7.v_proj": 441, | |
| "7.o_proj": 1110, | |
| "7.gate_proj": 2016, | |
| "7.up_proj": 2016, | |
| "7.down_proj": 1688, | |
| "8.q_proj": 1122, | |
| "8.k_proj": 444, | |
| "8.v_proj": 441, | |
| "8.o_proj": 1109, | |
| "8.gate_proj": 2016, | |
| "8.up_proj": 2016, | |
| "8.down_proj": 1685, | |
| "9.q_proj": 1121, | |
| "9.k_proj": 444, | |
| "9.v_proj": 442, | |
| "9.o_proj": 1109, | |
| "9.gate_proj": 2016, | |
| "9.up_proj": 2016, | |
| "9.down_proj": 1692, | |
| "10.q_proj": 1122, | |
| "10.k_proj": 444, | |
| "10.v_proj": 442, | |
| "10.o_proj": 1118, | |
| "10.gate_proj": 2016, | |
| "10.up_proj": 2016, | |
| "10.down_proj": 1692, | |
| "11.q_proj": 1120, | |
| "11.k_proj": 444, | |
| "11.v_proj": 443, | |
| "11.o_proj": 1121, | |
| "11.gate_proj": 2016, | |
| "11.up_proj": 2016, | |
| "11.down_proj": 1695, | |
| "12.q_proj": 1127, | |
| "12.k_proj": 447, | |
| "12.v_proj": 447, | |
| "12.o_proj": 1129, | |
| "12.gate_proj": 2016, | |
| "12.up_proj": 2016, | |
| "12.down_proj": 1714, | |
| "13.q_proj": 1123, | |
| "13.k_proj": 446, | |
| "13.v_proj": 445, | |
| "13.o_proj": 1127, | |
| "13.gate_proj": 2016, | |
| "13.up_proj": 2016, | |
| "13.down_proj": 1714, | |
| "14.q_proj": 1132, | |
| "14.k_proj": 451, | |
| "14.v_proj": 453, | |
| "14.o_proj": 1139, | |
| "14.gate_proj": 2016, | |
| "14.up_proj": 2016, | |
| "14.down_proj": 1742, | |
| "15.q_proj": 1145, | |
| "15.k_proj": 458, | |
| "15.v_proj": 461, | |
| "15.o_proj": 1157, | |
| "15.gate_proj": 2016, | |
| "15.up_proj": 2016, | |
| "15.down_proj": 1777, | |
| "16.q_proj": 1151, | |
| "16.k_proj": 461, | |
| "16.v_proj": 463, | |
| "16.o_proj": 1162, | |
| "16.gate_proj": 2016, | |
| "16.up_proj": 2016, | |
| "16.down_proj": 1800, | |
| "17.q_proj": 1174, | |
| "17.k_proj": 473, | |
| "17.v_proj": 476, | |
| "17.o_proj": 1188, | |
| "17.gate_proj": 2016, | |
| "17.up_proj": 2016, | |
| "17.down_proj": 1852, | |
| "18.q_proj": 1174, | |
| "18.k_proj": 472, | |
| "18.v_proj": 476, | |
| "18.o_proj": 1191, | |
| "18.gate_proj": 2016, | |
| "18.up_proj": 2016, | |
| "18.down_proj": 1866, | |
| "19.q_proj": 1158, | |
| "19.k_proj": 465, | |
| "19.v_proj": 469, | |
| "19.o_proj": 1179, | |
| "19.gate_proj": 2016, | |
| "19.up_proj": 2016, | |
| "19.down_proj": 1844, | |
| "20.q_proj": 1130, | |
| "20.k_proj": 449, | |
| "20.v_proj": 454, | |
| "20.o_proj": 1146, | |
| "20.gate_proj": 2016, | |
| "20.up_proj": 2016, | |
| "20.down_proj": 1786, | |
| "21.q_proj": 1123, | |
| "21.k_proj": 444, | |
| "21.v_proj": 451, | |
| "21.o_proj": 1137, | |
| "21.gate_proj": 2016, | |
| "21.up_proj": 2016, | |
| "21.down_proj": 1772, | |
| "22.q_proj": 1112, | |
| "22.k_proj": 437, | |
| "22.v_proj": 445, | |
| "22.o_proj": 1121, | |
| "22.gate_proj": 2016, | |
| "22.up_proj": 2016, | |
| "22.down_proj": 1749, | |
| "23.q_proj": 1112, | |
| "23.k_proj": 438, | |
| "23.v_proj": 444, | |
| "23.o_proj": 1129, | |
| "23.gate_proj": 2016, | |
| "23.up_proj": 2016, | |
| "23.down_proj": 1752, | |
| "24.q_proj": 1112, | |
| "24.k_proj": 437, | |
| "24.v_proj": 445, | |
| "24.o_proj": 1133, | |
| "24.gate_proj": 2016, | |
| "24.up_proj": 2016, | |
| "24.down_proj": 1754, | |
| "25.q_proj": 1114, | |
| "25.k_proj": 439, | |
| "25.v_proj": 449, | |
| "25.o_proj": 1141, | |
| "25.gate_proj": 2016, | |
| "25.up_proj": 2016, | |
| "25.down_proj": 1766, | |
| "26.q_proj": 1119, | |
| "26.k_proj": 441, | |
| "26.v_proj": 452, | |
| "26.o_proj": 1149, | |
| "26.gate_proj": 2016, | |
| "26.up_proj": 2016, | |
| "26.down_proj": 1780, | |
| "27.q_proj": 1123, | |
| "27.k_proj": 444, | |
| "27.v_proj": 454, | |
| "27.o_proj": 1149, | |
| "27.gate_proj": 2016, | |
| "27.up_proj": 2016, | |
| "27.down_proj": 1792, | |
| "28.q_proj": 1139, | |
| "28.k_proj": 452, | |
| "28.v_proj": 463, | |
| "28.o_proj": 1171, | |
| "28.gate_proj": 2016, | |
| "28.up_proj": 2016, | |
| "28.down_proj": 1838, | |
| "29.q_proj": 1148, | |
| "29.k_proj": 457, | |
| "29.v_proj": 469, | |
| "29.o_proj": 1193, | |
| "29.gate_proj": 2016, | |
| "29.up_proj": 2016, | |
| "29.down_proj": 1861, | |
| "30.q_proj": 1165, | |
| "30.k_proj": 466, | |
| "30.v_proj": 481, | |
| "30.o_proj": 1215, | |
| "30.gate_proj": 2016, | |
| "30.up_proj": 2016, | |
| "30.down_proj": 1912, | |
| "31.q_proj": 1263, | |
| "31.k_proj": 519, | |
| "31.v_proj": 536, | |
| "31.o_proj": 1336, | |
| "31.gate_proj": 2016, | |
| "31.up_proj": 2016, | |
| "31.down_proj": 2179 | |
| }, | |
| "rel_err": { | |
| "0.q_proj": 0.09311120212078094, | |
| "0.k_proj": 0.08253687620162964, | |
| "0.v_proj": 0.3433501720428467, | |
| "0.o_proj": 0.3911000192165375, | |
| "0.gate_proj": 0.3707573413848877, | |
| "0.up_proj": 0.41067802906036377, | |
| "0.down_proj": 0.5313096046447754, | |
| "1.q_proj": 0.16670213639736176, | |
| "1.k_proj": 0.10798212885856628, | |
| "1.v_proj": 0.5118529796600342, | |
| "1.o_proj": 0.5213813781738281, | |
| "1.gate_proj": 0.4368397891521454, | |
| "1.up_proj": 0.4705049991607666, | |
| "1.down_proj": 0.4971780776977539, | |
| "2.q_proj": 0.46672654151916504, | |
| "2.k_proj": 0.46059876680374146, | |
| "2.v_proj": 0.6164757013320923, | |
| "2.o_proj": 0.5800544619560242, | |
| "2.gate_proj": 0.512566328048706, | |
| "2.up_proj": 0.5365164875984192, | |
| "2.down_proj": 0.6227588057518005, | |
| "3.q_proj": 0.4239010214805603, | |
| "3.k_proj": 0.3750090003013611, | |
| "3.v_proj": 0.6444101929664612, | |
| "3.o_proj": 0.6232957243919373, | |
| "3.gate_proj": 0.5219470858573914, | |
| "3.up_proj": 0.551502525806427, | |
| "3.down_proj": 0.6144695281982422, | |
| "4.q_proj": 0.4298405945301056, | |
| "4.k_proj": 0.3984021842479706, | |
| "4.v_proj": 0.6045357584953308, | |
| "4.o_proj": 0.5685608386993408, | |
| "4.gate_proj": 0.523536205291748, | |
| "4.up_proj": 0.5611042976379395, | |
| "4.down_proj": 0.6211814284324646, | |
| "5.q_proj": 0.46155083179473877, | |
| "5.k_proj": 0.4410848319530487, | |
| "5.v_proj": 0.6360407471656799, | |
| "5.o_proj": 0.6054348349571228, | |
| "5.gate_proj": 0.5178220868110657, | |
| "5.up_proj": 0.5632009506225586, | |
| "5.down_proj": 0.6081773638725281, | |
| "6.q_proj": 0.4475008249282837, | |
| "6.k_proj": 0.4137071967124939, | |
| "6.v_proj": 0.6230500340461731, | |
| "6.o_proj": 0.5891891121864319, | |
| "6.gate_proj": 0.5309041738510132, | |
| "6.up_proj": 0.5662814378738403, | |
| "6.down_proj": 0.5977262258529663, | |
| "7.q_proj": 0.427289217710495, | |
| "7.k_proj": 0.38710105419158936, | |
| "7.v_proj": 0.6161573529243469, | |
| "7.o_proj": 0.5648438930511475, | |
| "7.gate_proj": 0.5202533602714539, | |
| "7.up_proj": 0.5593512654304504, | |
| "7.down_proj": 0.592329740524292, | |
| "8.q_proj": 0.43415379524230957, | |
| "8.k_proj": 0.3839348256587982, | |
| "8.v_proj": 0.576714813709259, | |
| "8.o_proj": 0.5553375482559204, | |
| "8.gate_proj": 0.523369312286377, | |
| "8.up_proj": 0.5563741326332092, | |
| "8.down_proj": 0.5858190059661865, | |
| "9.q_proj": 0.42894482612609863, | |
| "9.k_proj": 0.3965420722961426, | |
| "9.v_proj": 0.5610228180885315, | |
| "9.o_proj": 0.5573288202285767, | |
| "9.gate_proj": 0.5195508599281311, | |
| "9.up_proj": 0.5543780326843262, | |
| "9.down_proj": 0.5704837441444397, | |
| "10.q_proj": 0.42824241518974304, | |
| "10.k_proj": 0.401052862405777, | |
| "10.v_proj": 0.5595612525939941, | |
| "10.o_proj": 0.5597419738769531, | |
| "10.gate_proj": 0.5190359354019165, | |
| "10.up_proj": 0.5517367720603943, | |
| "10.down_proj": 0.5768575668334961, | |
| "11.q_proj": 0.4241175949573517, | |
| "11.k_proj": 0.3747548758983612, | |
| "11.v_proj": 0.5366345643997192, | |
| "11.o_proj": 0.5581367611885071, | |
| "11.gate_proj": 0.5184133052825928, | |
| "11.up_proj": 0.5504465103149414, | |
| "11.down_proj": 0.5751660466194153, | |
| "12.q_proj": 0.41172537207603455, | |
| "12.k_proj": 0.3747641444206238, | |
| "12.v_proj": 0.5728425979614258, | |
| "12.o_proj": 0.5404180288314819, | |
| "12.gate_proj": 0.5148531198501587, | |
| "12.up_proj": 0.5455449223518372, | |
| "12.down_proj": 0.5717161893844604, | |
| "13.q_proj": 0.4387548565864563, | |
| "13.k_proj": 0.40697500109672546, | |
| "13.v_proj": 0.5708914399147034, | |
| "13.o_proj": 0.5474628210067749, | |
| "13.gate_proj": 0.5218628644943237, | |
| "13.up_proj": 0.5454148054122925, | |
| "13.down_proj": 0.5676140189170837, | |
| "14.q_proj": 0.4094313979148865, | |
| "14.k_proj": 0.38163262605667114, | |
| "14.v_proj": 0.5798335075378418, | |
| "14.o_proj": 0.5494734048843384, | |
| "14.gate_proj": 0.5187147259712219, | |
| "14.up_proj": 0.5501715540885925, | |
| "14.down_proj": 0.593805193901062, | |
| "15.q_proj": 0.40906858444213867, | |
| "15.k_proj": 0.3624928295612335, | |
| "15.v_proj": 0.6080702543258667, | |
| "15.o_proj": 0.5543991327285767, | |
| "15.gate_proj": 0.5166524648666382, | |
| "15.up_proj": 0.5478196144104004, | |
| "15.down_proj": 0.5904906988143921, | |
| "16.q_proj": 0.4163205027580261, | |
| "16.k_proj": 0.3787270486354828, | |
| "16.v_proj": 0.6027750372886658, | |
| "16.o_proj": 0.5409226417541504, | |
| "16.gate_proj": 0.520693838596344, | |
| "16.up_proj": 0.556447446346283, | |
| "16.down_proj": 0.5904657244682312, | |
| "17.q_proj": 0.4439752697944641, | |
| "17.k_proj": 0.41725632548332214, | |
| "17.v_proj": 0.5800758004188538, | |
| "17.o_proj": 0.5581274628639221, | |
| "17.gate_proj": 0.5324439406394958, | |
| "17.up_proj": 0.5666811466217041, | |
| "17.down_proj": 0.5825478434562683, | |
| "18.q_proj": 0.4187489449977875, | |
| "18.k_proj": 0.3872785270214081, | |
| "18.v_proj": 0.6302375793457031, | |
| "18.o_proj": 0.5343031883239746, | |
| "18.gate_proj": 0.5309059619903564, | |
| "18.up_proj": 0.564786434173584, | |
| "18.down_proj": 0.5998350381851196, | |
| "19.q_proj": 0.4693866968154907, | |
| "19.k_proj": 0.40707793831825256, | |
| "19.v_proj": 0.6421195268630981, | |
| "19.o_proj": 0.5577782392501831, | |
| "19.gate_proj": 0.536626398563385, | |
| "19.up_proj": 0.5659677386283875, | |
| "19.down_proj": 0.6147719621658325, | |
| "20.q_proj": 0.4903647303581238, | |
| "20.k_proj": 0.44427719712257385, | |
| "20.v_proj": 0.6202300190925598, | |
| "20.o_proj": 0.5741354823112488, | |
| "20.gate_proj": 0.5429403185844421, | |
| "20.up_proj": 0.5663204789161682, | |
| "20.down_proj": 0.6225840449333191, | |
| "21.q_proj": 0.5229761004447937, | |
| "21.k_proj": 0.47562214732170105, | |
| "21.v_proj": 0.5926244854927063, | |
| "21.o_proj": 0.611431360244751, | |
| "21.gate_proj": 0.5432093143463135, | |
| "21.up_proj": 0.5681016445159912, | |
| "21.down_proj": 0.6183050274848938, | |
| "22.q_proj": 0.5626530051231384, | |
| "22.k_proj": 0.5008488297462463, | |
| "22.v_proj": 0.5983458757400513, | |
| "22.o_proj": 0.6302191019058228, | |
| "22.gate_proj": 0.5491480231285095, | |
| "22.up_proj": 0.5697852969169617, | |
| "22.down_proj": 0.6164385676383972, | |
| "23.q_proj": 0.551480233669281, | |
| "23.k_proj": 0.5012710690498352, | |
| "23.v_proj": 0.6238268613815308, | |
| "23.o_proj": 0.6080763339996338, | |
| "23.gate_proj": 0.5521300435066223, | |
| "23.up_proj": 0.5716087222099304, | |
| "23.down_proj": 0.6099771857261658, | |
| "24.q_proj": 0.5572198629379272, | |
| "24.k_proj": 0.5183809399604797, | |
| "24.v_proj": 0.6330755352973938, | |
| "24.o_proj": 0.6300069689750671, | |
| "24.gate_proj": 0.5530964136123657, | |
| "24.up_proj": 0.5734592080116272, | |
| "24.down_proj": 0.600332498550415, | |
| "25.q_proj": 0.5632444620132446, | |
| "25.k_proj": 0.5026987791061401, | |
| "25.v_proj": 0.6118474006652832, | |
| "25.o_proj": 0.6110818982124329, | |
| "25.gate_proj": 0.5554454922676086, | |
| "25.up_proj": 0.57483971118927, | |
| "25.down_proj": 0.588827908039093, | |
| "26.q_proj": 0.558449923992157, | |
| "26.k_proj": 0.4919636845588684, | |
| "26.v_proj": 0.5971678495407104, | |
| "26.o_proj": 0.5927439332008362, | |
| "26.gate_proj": 0.5562448501586914, | |
| "26.up_proj": 0.5749430060386658, | |
| "26.down_proj": 0.5749134421348572, | |
| "27.q_proj": 0.5691151022911072, | |
| "27.k_proj": 0.5340715050697327, | |
| "27.v_proj": 0.6143185496330261, | |
| "27.o_proj": 0.5845725536346436, | |
| "27.gate_proj": 0.5597459673881531, | |
| "27.up_proj": 0.5763874650001526, | |
| "27.down_proj": 0.5742335319519043, | |
| "28.q_proj": 0.5281823873519897, | |
| "28.k_proj": 0.4590943157672882, | |
| "28.v_proj": 0.5685400366783142, | |
| "28.o_proj": 0.5630350112915039, | |
| "28.gate_proj": 0.5594995617866516, | |
| "28.up_proj": 0.5738884806632996, | |
| "28.down_proj": 0.5813770294189453, | |
| "29.q_proj": 0.5108082890510559, | |
| "29.k_proj": 0.46057870984077454, | |
| "29.v_proj": 0.6014853715896606, | |
| "29.o_proj": 0.5444827079772949, | |
| "29.gate_proj": 0.5542632937431335, | |
| "29.up_proj": 0.5680719017982483, | |
| "29.down_proj": 0.5874636769294739, | |
| "30.q_proj": 0.4883614480495453, | |
| "30.k_proj": 0.4195691645145416, | |
| "30.v_proj": 0.535009503364563, | |
| "30.o_proj": 0.5329458713531494, | |
| "30.gate_proj": 0.5540768504142761, | |
| "30.up_proj": 0.5595806241035461, | |
| "30.down_proj": 0.5732670426368713, | |
| "31.q_proj": 0.4402003884315491, | |
| "31.k_proj": 0.3733157217502594, | |
| "31.v_proj": 0.48748257756233215, | |
| "31.o_proj": 0.45593205094337463, | |
| "31.gate_proj": 0.5609548091888428, | |
| "31.up_proj": 0.5358024835586548, | |
| "31.down_proj": 0.5254908800125122 | |
| }, | |
| "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": "shape", | |
| "param_fraction": 0.6003230168269231, | |
| "mean_rel_err": 0.523012746524598, | |
| "max_rel_err": 0.6436969041824341 | |
| }, | |
| "factorisation": { | |
| "max_rel_err": 0.001442187582142651, | |
| "factor_params": 4189847552, | |
| "dense_params": 6979321856 | |
| }, | |
| "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)" | |
| } | |
| } |