DeepSeek-V4-Flash-0731-CRACK-2x-DGX-Spark
Reproducible refusal-subspace editing and TP=2 deployment of DeepSeek V4 Flash 0731 on two DGX Sparks
File Explorer
- ci.yml
- attn-out-sra-r8.report.json
- attn-out-sra-r8.safetensors
- .gitkeep
- .gitkeep
- bilingual-behavior-v1.jsonl
- writing-v1.jsonl
- .gitkeep
- docker-compose.capture.yml
- production.env.example
- EVALUATION.md
- INTELLIGENCE_PRESERVATION.md
- METHOD.md
- PUBLISHING.md
- RELEASE_RESULTS.json
- RESEARCH.md
- VALIDATION.md
- hotfix-dsv4-adaptive-topk-50004.sh
- hotfix-dsv4-grammar-advance.sh
- hotfix-dsv4-issue26-hybrid-swa-min.py
- hotfix-dsv4-issue27-partial-prefill-concurrency.py
- hotfix-dsv4-issue31-v2-thinking-budget.py
- hotfix-dsv4-mtp-buffer-50312.sh
- hotfix-dsv4-skip-topk-49486.sh
- hotfix-dsv4-sm12x-prefill-topk-49897.sh
- hotfix-encoding-dsv4-issue21.py
- hotfix-nvfp4-ds-mla-issue22.sh
- dspark_proposer.py
- .gitkeep
- docker-compose.dspark.yml
- LICENSE.upstream
- start-deepseek-v4-flash-dspark.sh
- stop-deepseek-v4-flash-dspark.sh
- UPSTREAM.md
- sitecustomize.py
- __init__.py
- build_calibration_dataset.py
- build_capability_suite.py
- build_release_pair.sh
- capture_client.py
- compare_behavior.py
- compare_capability.py
- compare_judgments.py
- edit_checkpoint_in_runtime.sh
- evaluate_behavior.py
- evaluate_capability.py
- evaluate_gsm8k.py
- evaluate_humaneval.py
- evaluate_mmlu_pro.py
- evaluate_writing.py
- fetch_captures.sh
- fetch_eval_datasets.py
- judge_behavior.py
- judge_writing.py
- prepare_mmlu_pro.py
- score_humaneval.py
- score_humaneval_in_container.sh
- start_capture_cluster.sh
- start_production_cluster.sh
- stop_capture_cluster.sh
- stop_production_cluster.sh
- sync_to_sparks.sh
- validate_candidate_in_runtime.sh
- verify_source_in_runtime.sh
- __init__.py
- directions.py
- edit_checkpoint.py
- identity.py
- quantization.py
- validate_candidate.py
- test_bilingual_dataset.py
- test_compare_behavior.py
- test_compare_capability.py
- test_compare_judgments.py
- test_directions.py
- test_edit_checkpoint.py
- test_evaluate_capability.py
- test_evaluate_gsm8k.py
- test_evaluate_humaneval.py
- test_evaluate_mmlu_pro.py
- test_identity.py
- test_judge_behavior.py
- test_quantization.py
- test_release_artifact.py
- test_validate_candidate.py
- test_writing.py
- .env.example
- .gitattributes
- .gitignore
- LICENSE
- model-lock.json
- MODEL_CARD.md
- pyproject.toml
- README.md
- THIRD_PARTY_NOTICES.md
# Use via CDN
jsDelivrjsDelivr serves any public GitHub repository as a CDN with zero setup. Pick a version and a file to get a ready-to-paste link and snippet.
Command Glossary
Commands referenced in this DOCs, explained below.
pytest
View Details ▼
pytest
Run Python tests.
pytest {{path/to/test_file1.py path/to/test_file2.py ...}}
Run tests from specific files:
pytest -k {{expression}}
Run tests with names matching a specific [k]eyword expression:
pytest {{[-x|--exitfirst]}}
Exit as soon as a test fails or encounters an error:
python
View Details ▼
python
Python language interpreter.
python
Start a REPL (interactive shell):
python {{path/to/file.py}}
Execute a specific Python file:
python -i {{path/to/file.py}}
Execute a specific Python file and start a REPL:
ruff check
View Details ▼
ruff check
An extremely fast Python linter. `check` is the default command - it can be omitted everywhere.
If no files or directories are specified, the current working directory is used by default.
ruff check {{path/to/file_or_directory1 path/to/file_or_directory2 ...}}
Run the linter on the given files or directories:
ruff check --fix
Apply the suggested fixes, modifying the files in-place:
ruff check --watch
Run the linter and re-lint on change:
