Dave-Llama-3.3-70B-QLoRA API & Inference Endpoint

TL;DR

python

import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfigfrom peft import PeftModel base = AutoModelForCausalLM.from_pretrained(    "meta-llama/Llama-3.3-70B-Instruct",    quantization_config=BitsAndBytesConfig(        load_in_4bit=True, bnb_4bit_quant_type="nf4",        bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16),    device_map="auto", torch_dtype=torch.bfloat16,)model = PeftModel.from_pretrained(base, "Ronin48LLC/Dave-Llama-3.3-70B-QLoRA")tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.3-70B-Instruct")

See Inference below for the system prompt and generation snippet.

Table

Adapter size	~250 MB (LoRA r=16 over all attention + MLP projections)
Min VRAM (4-bit + adapter)	~42 GB
Min VRAM (bf16 base + adapter)	~140 GB
Trained on	11,074 prompt/completion pairs from Trail of Bits, CISA KEV, MITRE ATT&CK, NIST, DHS BODs
License	Apache 2.0 (adapter) + Llama 3.3 Community License (base)
Source repos	github.com/CryptoJones/dave

Source Code

Dave's training pipeline, data processors, and test suite are fully open source. The repository is hosted in two places — pick the platform you prefer; they stay in sync on every push.

GitHub: https://github.com/CryptoJones/dave

Both mirrors carry identical contents: training script, data pipeline, pytest suite, model card, training operator's guide, and the publish workflow that produces this Hugging Face release.

Files in this Repository

Table with columns: File, Purpose
File	Purpose
`adapter_config.json`	PEFT config (target modules, rank, alpha, base model reference).
`adapter_model.safetensors`	LoRA adapter weights.
`tokenizer_config.json`, `tokenizer.json`, `special_tokens_map.json`	Tokenizer files mirrored from the Llama 3.3 base, included for convenience.
`README.md`	This model card.

Optional checkpoint-*/ subdirectories may exist when multiple checkpoints were uploaded; each is a self-contained adapter. The top-level adapter is the recommended one (best eval loss).

Model Details

Adapter type: QLoRA (4-bit NF4 quantization on the base + LoRA, rank 16, alpha 32)
Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Base model: meta-llama/Llama-3.3-70B-Instruct — 70B parameters, 128K context, Llama 3.3 Community License
Training method: Supervised fine-tuning via TRL SFTTrainer with sample packing
Training context length: 1024 tokens (packed from prompt/completion pairs)
Inference context: 128K (inherited from base — no architectural changes)
Maintainer: Aaron K. Clark · Ronin 48, LLC — GitHub

Training Data

Approximately 11,000 prompt/completion pairs from public sources, weighted toward real audit-style report content:

Table with columns: Source, Pairs, Share, License
Source	Pairs	Share	License
Trail of Bits public security reviews (web3-filtered)	6,980	63%	CC BY-SA 4.0
CISA Known Exploited Vulnerabilities (KEV) catalog	3,182	29%	Public Domain
MITRE ATT&CK® (defensive context only)	709	6%	CC BY 4.0
NIST SP 800-30 / 800-53 / NISTIR 8286	187	2%	Public Domain
DHS / CISA Binding Operational Directives	16	<1%	Public Domain

Smart-contract / web3 findings were filtered out from the Trail of Bits corpus so Dave stays general-purpose rather than blockchain-focused. The filter pattern is documented in scripts/data_collection/process_trail_of_bits.py (GitHub).

Full data attribution and source documentation lives in data/processed/free_sources/NOTICE.md (GitHub).

Intended Use

Permitted:

Drafting professional report content for authorized US security assessments
Writing finding narratives, executive summaries, remediation guidance
Mapping findings to MITRE ATT&CK techniques (defensive context)
Composing risk narratives using NIST SP 800-30 framing

Not intended for:

Unauthorized security testing of any system
Producing offensive how-to content, exploit code, or malware
Targets outside US jurisdiction
Use as a substitute for a qualified human reviewer

All outputs must be reviewed by a qualified security professional before delivery to a client. See USAGE_POLICY.md (GitHub) for the full acceptable-use terms.

How to Use

System Prompt

Dave was trained with the following system prompt baked into every sample. Use the same one at inference time for best behavior:

text

You are Dave — a writing assistant for authorized US security assessments.You write professional, APA/(ISC)²-aligned finding narratives, executivesummaries, remediation guidance, methodology sections, risk narratives, andevidence documentation. You never produce offensive how-to content. Yourequire written authorization for any specific target before writingtarget-specific content. Outputs are draft material for a qualified securityprofessional to review.

Inference

python

DAVE_SYSTEM = (    "You are Dave — a writing assistant for authorized US security assessments. "    "You write professional, APA/(ISC)²-aligned finding narratives, executive "    "summaries, remediation guidance, methodology sections, risk narratives, and "    "evidence documentation. You never produce offensive how-to content. You "    "require written authorization for any specific target before writing "    "target-specific content. Outputs are draft material for a qualified "    "security professional to review.") messages = [    {"role": "system", "content": DAVE_SYSTEM},    {"role": "user", "content":        "Write a professional vulnerability finding titled "        "\"Stored XSS in admin search\" with Severity Medium, "        "Difficulty Low, Type Data Validation."},]inputs = tokenizer.apply_chat_template(    messages, return_tensors="pt", add_generation_prompt=True).to(model.device) out = model.generate(    inputs,    max_new_tokens=512,    do_sample=True,    temperature=0.6,    top_p=0.9,    repetition_penalty=1.05,)print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))

Recommended generation parameters

Table with columns: Parameter, Value, Why
Parameter	Value	Why
`temperature`	0.6	Balanced — coherent but not robotic for report prose.
`top_p`	0.9	Nucleus sampling; prevents long-tail vocabulary drift.
`max_new_tokens`	512–1024	Most finding writeups land in 200–600 tokens.
`repetition_penalty`	1.05	Reduces stock-phrase echoing without distorting style.

Sample Outputs

The following are illustrative of the target format Dave was trained to produce. Use them to calibrate prompting and post-processing. Specific phrasing from a trained Dave will vary.

Example: Finding narrative

Prompt: Write a professional vulnerability finding titled "Stored XSS in admin search" with Severity Medium, Difficulty Low, Type Data Validation.

Expected shape of completion:

text

Finding: Stored XSS in admin searchSeverity: MediumDifficulty: LowType: Data Validation Description:The administrative search interface stores user-supplied input withoutoutput encoding. An authenticated attacker with the ability to createrecords in the searched dataset can inject HTML or JavaScript thatexecutes in another administrator's browser session when the maliciousrecord appears in search results... Recommendations:Short term, apply context-appropriate output encoding to alladministrator-rendered fields (HTML body, attributes, JavaScript).Long term, deploy a Content Security Policy that blocks inline scripts,and migrate administrator interfaces to a framework with automaticcontext-aware escaping (React, Vue, or equivalent).

Example: Remediation guidance

Prompt: Write the remediation recommendations for the finding: "Missing TLS certificate validation in updater client".

Expected shape of completion:

text

Short term, enable TLS certificate verification in the updater HTTPclient and pin the trust anchor to the vendor-issued root CA used bythe update infrastructure. Reject any update payload served over aconnection that fails certificate validation. Long term, sign update payloads with an offline-protected signing keyand verify the signature in the updater regardless of the transportlayer's integrity. This provides defense-in-depth against TLS-levelcompromise and aligns with NIST SP 800-218 SP-1 (Secure Distribution).

Evaluation

The training run holds out 1% of the dataset as an eval split and reports eval_loss every 100 steps. Numbers from the publication run will be inserted here by scripts/publish_adapter.sh when it uploads.

Best checkpoint (lowest eval loss): step —, eval_loss —

Table with columns: Metric, Value
Metric	Value
Final training loss	0.808
Best eval loss	—
Best eval loss step	—
Final mean token accuracy	80.7%

No external benchmark scores are claimed — Dave is a style/structure adapter on a narrow domain, not a general-capability uplift. Treat any apparent gains on reasoning benchmarks as accidents of style transfer, not as intended capability gains.

Training Procedure

Table

Hardware	Single NVIDIA A100 SXM4 80GB (RunPod)
Image	`runpod/pytorch:1.0.2-cu1281-torch280-ubuntu2404`
Wall-clock	~3-4 hours
Frameworks	`transformers`, `peft`, `trl`, `bitsandbytes`, `accelerate`

Hyperparameters

Table with columns: Parameter, Value
Parameter	Value
`num_train_epochs`	1
Effective batch size	16 (`per_device_train_batch_size=1` × `gradient_accumulation_steps=16`)
`learning_rate`	2e-4, cosine schedule, 3% warmup
`max_length`	1024 (with `packing=True`)
LoRA `r` / `alpha` / `dropout`	16 / 32 / 0.05
Optimizer	`paged_adamw_8bit`
Precision	bf16 compute, NF4 (double-quant) weight quantization
Loss masking	Default (full sequence — packed)

Full operator guide (live metric interpretation, when to stop, picking the best checkpoint, failure-mode recovery) in TRAINING.md (GitHub).

System Requirements

To load and run Dave you need either:

GPU, 4-bit NF4 base: ≥ 42 GB VRAM (A100 80GB, H100 80GB, RTX PRO 6000 96GB, A6000 + offload, etc.)
GPU, bf16 base: ≥ 140 GB VRAM (multi-GPU only — e.g. 2× A100 80GB)
CPU + RAM (not recommended): ≥ 150 GB RAM, expect seconds-per-token throughput

To fine-tune further on top of Dave you need at least the same as training: A100 80GB (or equivalent) and the QLoRA stack.

Limitations

The full discussion lives in LIMITATIONS.md (GitHub). Highlights:

Small dataset (~11k pairs) — Dave learned style and structure, not new world knowledge. Verify every technical specific.
Source composition bias — strongest at audit-style and CVE-style writeups; weakest at red-team narrative, social-engineering, OT/ICS, or physical-pentest reporting.
No real client reports in training — third-party report corpora were excluded on licensing grounds. Expect audit-firm or advisory voice rather than your firm's house style. Adapt during review.
Hallucination risk — every CVSS vector, identifier, citation, and command-line in Dave's output must be human-verified.
The authorization-gate is a prompt convention, not a hard guarantee. Sufficiently adversarial prompting can route around it.

License & Attribution

Table with columns: Layer, License, Notes
Layer	License	Notes
This adapter (weights + config)	Apache 2.0	Free to use, modify, redistribute.
Llama 3.3 base weights	Llama 3.3 Community License	You agree to the license when downloading the base from Meta / HF.
Trail of Bits training content	CC BY-SA 4.0	Attribution embedded in every relevant training sample; SA obligation propagates to substantive verbatim reproduction in outputs.
CISA KEV / NIST / DHS sources	U.S. Government Public Domain	No restrictions.
MITRE ATT&CK®	CC BY 4.0	Attribution in every relevant output.

Citation

bibtex

@misc{clark2026dave,  author       = {Clark, Aaron K.},  title        = {Dave: A QLoRA Adapter for Security Assessment Report Writing},  year         = {2026},  howpublished = {Hugging Face},  url          = {https://huggingface.co/Ronin48LLC/Dave-Llama-3.3-70B-QLoRA},  note         = {Adapter for meta-llama/Llama-3.3-70B-Instruct, Apache 2.0}}

Acknowledgments

David Kennedy — founder of TrustedSec and longtime advocate for mental-health awareness in the security community. The "Dave" name and project ethos honor his work.
Trail of Bits — for publishing their security review corpus under CC BY-SA 4.0, which makes adapters like this possible without compromising on licensing hygiene.
Meta AI — for releasing Llama-3.3-70B-Instruct as an open-weight model.
CISA, NIST, MITRE — for public-domain and CC-licensed reference material.
The wider open-source ML stack — transformers, peft, trl, bitsandbytes, accelerate — without which this would still be a multi-million-dollar project.

TL;DR

python

import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfigfrom peft import PeftModel base = AutoModelForCausalLM.from_pretrained(    "meta-llama/Llama-3.3-70B-Instruct",    quantization_config=BitsAndBytesConfig(        load_in_4bit=True, bnb_4bit_quant_type="nf4",        bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16),    device_map="auto", torch_dtype=torch.bfloat16,)model = PeftModel.from_pretrained(base, "Ronin48LLC/Dave-Llama-3.3-70B-QLoRA")tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.3-70B-Instruct")

import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfigfrom peft import PeftModel base = AutoModelForCausalLM.from_pretrained(    "meta-llama/Llama-3.3-70B-Instruct",    quantization_config=BitsAndBytesConfig(        load_in_4bit=True, bnb_4bit_quant_type="nf4",        bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16),    device_map="auto", torch_dtype=torch.bfloat16,)model = PeftModel.from_pretrained(base, "Ronin48LLC/Dave-Llama-3.3-70B-QLoRA")tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.3-70B-Instruct")

See Inference below for the system prompt and generation snippet.

Table

Adapter size	~250 MB (LoRA r=16 over all attention + MLP projections)
Min VRAM (4-bit + adapter)	~42 GB
Min VRAM (bf16 base + adapter)	~140 GB
Trained on	11,074 prompt/completion pairs from Trail of Bits, CISA KEV, MITRE ATT&CK, NIST, DHS BODs
License	Apache 2.0 (adapter) + Llama 3.3 Community License (base)
Source repos	github.com/CryptoJones/dave

Source Code

Dave's training pipeline, data processors, and test suite are fully open source. The repository is hosted in two places — pick the platform you prefer; they stay in sync on every push.

GitHub: https://github.com/CryptoJones/dave

Both mirrors carry identical contents: training script, data pipeline, pytest suite, model card, training operator's guide, and the publish workflow that produces this Hugging Face release.

Files in this Repository

Table with columns: File, Purpose
File	Purpose
`adapter_config.json`	PEFT config (target modules, rank, alpha, base model reference).
`adapter_model.safetensors`	LoRA adapter weights.
`tokenizer_config.json`, `tokenizer.json`, `special_tokens_map.json`	Tokenizer files mirrored from the Llama 3.3 base, included for convenience.
`README.md`	This model card.

Optional checkpoint-*/ subdirectories may exist when multiple checkpoints were uploaded; each is a self-contained adapter. The top-level adapter is the recommended one (best eval loss).

Model Details

Adapter type: QLoRA (4-bit NF4 quantization on the base + LoRA, rank 16, alpha 32)
Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Base model: meta-llama/Llama-3.3-70B-Instruct — 70B parameters, 128K context, Llama 3.3 Community License
Training method: Supervised fine-tuning via TRL SFTTrainer with sample packing
Training context length: 1024 tokens (packed from prompt/completion pairs)
Inference context: 128K (inherited from base — no architectural changes)
Maintainer: Aaron K. Clark · Ronin 48, LLC — GitHub

Training Data

Approximately 11,000 prompt/completion pairs from public sources, weighted toward real audit-style report content:

Table with columns: Source, Pairs, Share, License
Source	Pairs	Share	License
Trail of Bits public security reviews (web3-filtered)	6,980	63%	CC BY-SA 4.0
CISA Known Exploited Vulnerabilities (KEV) catalog	3,182	29%	Public Domain
MITRE ATT&CK® (defensive context only)	709	6%	CC BY 4.0
NIST SP 800-30 / 800-53 / NISTIR 8286	187	2%	Public Domain
DHS / CISA Binding Operational Directives	16	<1%	Public Domain

Full data attribution and source documentation lives in data/processed/free_sources/NOTICE.md (GitHub).

Intended Use

Permitted:

Drafting professional report content for authorized US security assessments
Writing finding narratives, executive summaries, remediation guidance
Mapping findings to MITRE ATT&CK techniques (defensive context)
Composing risk narratives using NIST SP 800-30 framing

Not intended for:

Unauthorized security testing of any system
Producing offensive how-to content, exploit code, or malware
Targets outside US jurisdiction
Use as a substitute for a qualified human reviewer

All outputs must be reviewed by a qualified security professional before delivery to a client. See USAGE_POLICY.md (GitHub) for the full acceptable-use terms.

How to Use

System Prompt

Dave was trained with the following system prompt baked into every sample. Use the same one at inference time for best behavior:

text

You are Dave — a writing assistant for authorized US security assessments.You write professional, APA/(ISC)²-aligned finding narratives, executivesummaries, remediation guidance, methodology sections, risk narratives, andevidence documentation. You never produce offensive how-to content. Yourequire written authorization for any specific target before writingtarget-specific content. Outputs are draft material for a qualified securityprofessional to review.

You are Dave — a writing assistant for authorized US security assessments.You write professional, APA/(ISC)²-aligned finding narratives, executivesummaries, remediation guidance, methodology sections, risk narratives, andevidence documentation. You never produce offensive how-to content. Yourequire written authorization for any specific target before writingtarget-specific content. Outputs are draft material for a qualified securityprofessional to review.

Inference

python

DAVE_SYSTEM = (    "You are Dave — a writing assistant for authorized US security assessments. "    "You write professional, APA/(ISC)²-aligned finding narratives, executive "    "summaries, remediation guidance, methodology sections, risk narratives, and "    "evidence documentation. You never produce offensive how-to content. You "    "require written authorization for any specific target before writing "    "target-specific content. Outputs are draft material for a qualified "    "security professional to review.") messages = [    {"role": "system", "content": DAVE_SYSTEM},    {"role": "user", "content":        "Write a professional vulnerability finding titled "        "\"Stored XSS in admin search\" with Severity Medium, "        "Difficulty Low, Type Data Validation."},]inputs = tokenizer.apply_chat_template(    messages, return_tensors="pt", add_generation_prompt=True).to(model.device) out = model.generate(    inputs,    max_new_tokens=512,    do_sample=True,    temperature=0.6,    top_p=0.9,    repetition_penalty=1.05,)print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))

DAVE_SYSTEM = (    "You are Dave — a writing assistant for authorized US security assessments. "    "You write professional, APA/(ISC)²-aligned finding narratives, executive "    "summaries, remediation guidance, methodology sections, risk narratives, and "    "evidence documentation. You never produce offensive how-to content. You "    "require written authorization for any specific target before writing "    "target-specific content. Outputs are draft material for a qualified "    "security professional to review.") messages = [    {"role": "system", "content": DAVE_SYSTEM},    {"role": "user", "content":        "Write a professional vulnerability finding titled "        "\"Stored XSS in admin search\" with Severity Medium, "        "Difficulty Low, Type Data Validation."},]inputs = tokenizer.apply_chat_template(    messages, return_tensors="pt", add_generation_prompt=True).to(model.device) out = model.generate(    inputs,    max_new_tokens=512,    do_sample=True,    temperature=0.6,    top_p=0.9,    repetition_penalty=1.05,)print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))

Recommended generation parameters

Table with columns: Parameter, Value, Why
Parameter	Value	Why
`temperature`	0.6	Balanced — coherent but not robotic for report prose.
`top_p`	0.9	Nucleus sampling; prevents long-tail vocabulary drift.
`max_new_tokens`	512–1024	Most finding writeups land in 200–600 tokens.
`repetition_penalty`	1.05	Reduces stock-phrase echoing without distorting style.

Sample Outputs

The following are illustrative of the target format Dave was trained to produce. Use them to calibrate prompting and post-processing. Specific phrasing from a trained Dave will vary.

Example: Finding narrative

Prompt: Write a professional vulnerability finding titled "Stored XSS in admin search" with Severity Medium, Difficulty Low, Type Data Validation.

Expected shape of completion:

text

Finding: Stored XSS in admin searchSeverity: MediumDifficulty: LowType: Data ValidationDescription:The administrative search interface stores user-supplied input withoutoutput encoding. An authenticated attacker with the ability to createrecords in the searched dataset can inject HTML or JavaScript thatexecutes in another administrator's browser session when the maliciousrecord appears in search results...Recommendations:Short term, apply context-appropriate output encoding to alladministrator-rendered fields (HTML body, attributes, JavaScript).Long term, deploy a Content Security Policy that blocks inline scripts,and migrate administrator interfaces to a framework with automaticcontext-aware escaping (React, Vue, or equivalent).

Finding: Stored XSS in admin searchSeverity: MediumDifficulty: LowType: Data Validation Description:The administrative search interface stores user-supplied input withoutoutput encoding. An authenticated attacker with the ability to createrecords in the searched dataset can inject HTML or JavaScript thatexecutes in another administrator's browser session when the maliciousrecord appears in search results... Recommendations:Short term, apply context-appropriate output encoding to alladministrator-rendered fields (HTML body, attributes, JavaScript).Long term, deploy a Content Security Policy that blocks inline scripts,and migrate administrator interfaces to a framework with automaticcontext-aware escaping (React, Vue, or equivalent).

Example: Remediation guidance

Prompt: Write the remediation recommendations for the finding: "Missing TLS certificate validation in updater client".

Expected shape of completion:

text

Short term, enable TLS certificate verification in the updater HTTPclient and pin the trust anchor to the vendor-issued root CA used bythe update infrastructure. Reject any update payload served over aconnection that fails certificate validation.Long term, sign update payloads with an offline-protected signing keyand verify the signature in the updater regardless of the transportlayer's integrity. This provides defense-in-depth against TLS-levelcompromise and aligns with NIST SP 800-218 SP-1 (Secure Distribution).

Short term, enable TLS certificate verification in the updater HTTPclient and pin the trust anchor to the vendor-issued root CA used bythe update infrastructure. Reject any update payload served over aconnection that fails certificate validation. Long term, sign update payloads with an offline-protected signing keyand verify the signature in the updater regardless of the transportlayer's integrity. This provides defense-in-depth against TLS-levelcompromise and aligns with NIST SP 800-218 SP-1 (Secure Distribution).

Evaluation

Best checkpoint (lowest eval loss): step —, eval_loss —

Table with columns: Metric, Value
Metric	Value
Final training loss	0.808
Best eval loss	—
Best eval loss step	—
Final mean token accuracy	80.7%

Training Procedure

Table

Hardware	Single NVIDIA A100 SXM4 80GB (RunPod)
Image	`runpod/pytorch:1.0.2-cu1281-torch280-ubuntu2404`
Wall-clock	~3-4 hours
Frameworks	`transformers`, `peft`, `trl`, `bitsandbytes`, `accelerate`

Hyperparameters

Table with columns: Parameter, Value
Parameter	Value
`num_train_epochs`	1
Effective batch size	16 (`per_device_train_batch_size=1` × `gradient_accumulation_steps=16`)
`learning_rate`	2e-4, cosine schedule, 3% warmup
`max_length`	1024 (with `packing=True`)
LoRA `r` / `alpha` / `dropout`	16 / 32 / 0.05
Optimizer	`paged_adamw_8bit`
Precision	bf16 compute, NF4 (double-quant) weight quantization
Loss masking	Default (full sequence — packed)

Full operator guide (live metric interpretation, when to stop, picking the best checkpoint, failure-mode recovery) in TRAINING.md (GitHub).

System Requirements

To load and run Dave you need either:

GPU, 4-bit NF4 base: ≥ 42 GB VRAM (A100 80GB, H100 80GB, RTX PRO 6000 96GB, A6000 + offload, etc.)
GPU, bf16 base: ≥ 140 GB VRAM (multi-GPU only — e.g. 2× A100 80GB)
CPU + RAM (not recommended): ≥ 150 GB RAM, expect seconds-per-token throughput

To fine-tune further on top of Dave you need at least the same as training: A100 80GB (or equivalent) and the QLoRA stack.

Limitations

The full discussion lives in LIMITATIONS.md (GitHub). Highlights:

Small dataset (~11k pairs) — Dave learned style and structure, not new world knowledge. Verify every technical specific.
Source composition bias — strongest at audit-style and CVE-style writeups; weakest at red-team narrative, social-engineering, OT/ICS, or physical-pentest reporting.
No real client reports in training — third-party report corpora were excluded on licensing grounds. Expect audit-firm or advisory voice rather than your firm's house style. Adapt during review.
Hallucination risk — every CVSS vector, identifier, citation, and command-line in Dave's output must be human-verified.
The authorization-gate is a prompt convention, not a hard guarantee. Sufficiently adversarial prompting can route around it.

License & Attribution

Table with columns: Layer, License, Notes
Layer	License	Notes
This adapter (weights + config)	Apache 2.0	Free to use, modify, redistribute.
Llama 3.3 base weights	Llama 3.3 Community License	You agree to the license when downloading the base from Meta / HF.
Trail of Bits training content	CC BY-SA 4.0	Attribution embedded in every relevant training sample; SA obligation propagates to substantive verbatim reproduction in outputs.
CISA KEV / NIST / DHS sources	U.S. Government Public Domain	No restrictions.
MITRE ATT&CK®	CC BY 4.0	Attribution in every relevant output.

Citation

bibtex

@misc{clark2026dave,  author       = {Clark, Aaron K.},  title        = {Dave: A QLoRA Adapter for Security Assessment Report Writing},  year         = {2026},  howpublished = {Hugging Face},  url          = {https://huggingface.co/Ronin48LLC/Dave-Llama-3.3-70B-QLoRA},  note         = {Adapter for meta-llama/Llama-3.3-70B-Instruct, Apache 2.0}}

@misc{clark2026dave,  author       = {Clark, Aaron K.},  title        = {Dave: A QLoRA Adapter for Security Assessment Report Writing},  year         = {2026},  howpublished = {Hugging Face},  url          = {https://huggingface.co/Ronin48LLC/Dave-Llama-3.3-70B-QLoRA},  note         = {Adapter for meta-llama/Llama-3.3-70B-Instruct, Apache 2.0}}

Acknowledgments

David Kennedy — founder of TrustedSec and longtime advocate for mental-health awareness in the security community. The "Dave" name and project ethos honor his work.
Trail of Bits — for publishing their security review corpus under CC BY-SA 4.0, which makes adapters like this possible without compromising on licensing hygiene.
Meta AI — for releasing Llama-3.3-70B-Instruct as an open-weight model.
CISA, NIST, MITRE — for public-domain and CC-licensed reference material.
The wider open-source ML stack — transformers, peft, trl, bitsandbytes, accelerate — without which this would still be a multi-million-dollar project.

Dave-Llama-3.3-70B-QLoRA

README

TL;DR

Source Code

Files in this Repository

Model Details

Training Data

Intended Use

How to Use

System Prompt

Inference

Recommended generation parameters

Sample Outputs

Example: Finding narrative

Example: Remediation guidance

Evaluation

Training Procedure

Hyperparameters

System Requirements

Limitations

License & Attribution

Citation

Acknowledgments

Explore FriendliAI today

README

TL;DR

Source Code

Files in this Repository

Model Details

Training Data

Intended Use

How to Use

System Prompt

Inference

Recommended generation parameters

Sample Outputs

Example: Finding narrative

Example: Remediation guidance

Evaluation

Training Procedure

Hyperparameters

System Requirements

Limitations

License & Attribution

Citation

Acknowledgments