Occultron Model Index

llama-3.1-8b-abliterated

manifest sha256:999b7f31c5cb71d078a7fc72acf6377880d8d7ad3af922aff5e5bb03c0a2fd7a · pricing $0.49/M · cached in $0.01/M · JSON

Nutrition label

artifact_format	unknown
artifact_provenance	unknown
artifact_verified	unverified
attestation	unknown
capability_bench	unknown
chat_template_sha256	unknown
checkpoint	llama-3.1-8b-abliterated
context	unknown
context_advertised	unknown
context_tested	unknown
decoding	unknown
derivative_revision	unknown
engine	vllm
gpu_arch	unknown
jurisdiction	unknown
license	unknown
moe	unknown
quant	unknown
quantization	unknown
refusal_bench	unknown
retention	none
runtime	vllm
runtime_version	unknown
sampler	unknown
seed_policy	unknown
sha256	unknown
throughput_tokens_per_sec	unknown
tokenizer_sha256	unknown
tool_use	unknown
ttft_ms_p50	unknown
upstream_repository	llama-3.1-8b-abliterated
upstream_revision	unknown
weight_manifest_root	unknown

Providers

id	class	assurance	attestation	healthy	cost $/M	attests manifest
local	wholesale	R1	none	True	0.0	False

Benchmarks

suite	score	version · date	metrics
Not yet benchmarked

AI Act Art.50 disclosure

schema eu-ai-act-art50/0.1 · digest sha256:79d00f8bf126569eed1da0a3601c68a58592fb572a4517addbd102e5cadc2cc6

ai_system	True
system_purpose	General-purpose text (and image) generation served over an OpenAI-compatible API. Output is machine-generated and may be inaccurate; it is not verified fact.
provider	Occultron
model_ref	llama-3.1-8b-abliterated
manifest_digest	sha256:999b7f31c5cb71d078a7fc72acf6377880d8d7ad3af922aff5e5bb03c0a2fd7a
generative	True
synthetic_content_marking	{'marked': True, 'method': 'response field `occultron.synthetic=true` + `x-ai-generated: true` header', 'provenance_claim': False, 'note': 'Output is marked as AI-generated; the mark does not assert the output is true, accurate or human-reviewed.'}
retention	none
jurisdiction
human_oversight	{'human_in_the_loop': False, 'note': 'Fully automated inference; the buyer is the deploying entity.'}
contact

Known limitations

Model output is probabilistic and can be false or misleading.
No content moderation beyond a narrow CSAM/mass-violence safety floor.
Receipts prove *which model executed*, not that its output is correct.

Canonical manifest

{
  "artifact": {
    "chat_template_sha256": null,
    "format": null,
    "merkle_root_sha256": null,
    "provenance": null,
    "tokenizer_sha256": null
  },
  "deployment": {
    "attestation": null,
    "container_digest": null,
    "jurisdiction": null,
    "moe": null,
    "quantization": null,
    "runtime": "vllm",
    "runtime_version": null
  },
  "determinism": {
    "decoding": null,
    "gpu_arch": null,
    "sampler": null,
    "seed_policy": null
  },
  "model_id": "llama-3.1-8b-abliterated",
  "policy": {
    "hidden_system_prompt": false,
    "middleware_transform": false
  },
  "revision": null,
  "schema": "model-manifest/0.1",
  "upstream": {
    "license": null,
    "repository": "llama-3.1-8b-abliterated"
  }
}

Machine-readable: /v1/models · verify receipts at /v1/receipts/verify. Benchmark scores are bound to a manifest digest, benchmark version and date; a model that changes revision does not inherit old scores.