{
  "model": "LiquidVamp-R4-1.7B",
  "author": "Wamphyre AI",
  "base_model": "orcarouter/LFM3-1.7B-abliterated",
  "teacher": "GLM-5.3-Flash",
  "modifications": [
    "Inherited neuronal-level FFN weight edits",
    "Supervised LoRA post-training, merged into weights"
  ],
  "training": {
    "teacher_unique_examples": 2163,
    "train_unique_examples": 1362,
    "validation_examples": 413,
    "heldout_examples": 388,
    "balanced_epoch_rows": 1611,
    "trainable_lora_parameters": 9999872,
    "epochs": 1,
    "optimizer_steps": 101,
    "precision": "float32",
    "hardware": "NVIDIA Tesla T4",
    "training_loss": 0.6319365055844335,
    "validation_loss": 0.5369187593460083,
    "assistant_only_loss": true,
    "all_final_run_gradients_finite": true
  },
  "evaluation": {
    "scope": "Local paired evaluation; correlated exercise families; not an external standard benchmark",
    "engine": "llama.cpp b12179 Vulkan",
    "hardware": "AMD Radeon RX 6600",
    "context": 8192,
    "temperature": 0,
    "top_k": 1,
    "repeat_penalty": 1,
    "speculative_decoding": false,
    "thinking": false,
    "cases_per_model": 436,
    "LiquidVamp_R4": {
      "heldout": {
        "passed": 357,
        "total": 388
      },
      "heldout_python": {
        "passed": 44,
        "total": 58
      },
      "heldout_reasoning": {
        "passed": 112,
        "total": 129
      },
      "heldout_simulated_tools": {
        "passed": 201,
        "total": 201
      },
      "historical_regressions": {
        "passed": 12,
        "total": 16
      },
      "fresh_transfer": {
        "passed": 13,
        "total": 32
      },
      "all": {
        "passed": 382,
        "total": 436
      }
    },
    "generation_retries": 0,
    "limitations": [
      "Python scores depend partly on executable code formatting",
      "Fresh transfer score is 13/32",
      "JavaScript all-block score is 1/5",
      "Tool observations in the heldout benchmark were simulated",
      "Long-context quality at 128K was not evaluated",
      "The OrcaRouter baseline has not been evaluated on this protocol"
    ],
    "comparison_baseline": {
      "model": "Wamphyre AI modified R3 predecessor",
      "status": "evaluated",
      "is_untouched_upstream_orcarouter": false
    },
    "comparative_gain_vs_baseline": null,
    "score_interpretation": "The measured comparison is modified R3 versus LiquidVamp R4. No untouched upstream OrcaRouter scores are claimed or estimated.",
    "numeric_punctuation_grader_corrected": true,
    "R3_modified_predecessor": {
      "identity": "OrcaRouter-derived experimental predecessor with prior Wamphyre AI modifications",
      "upstream_stock_model": false,
      "heldout": {
        "passed": 318,
        "total": 388
      },
      "heldout_python": {
        "passed": 9,
        "total": 58
      },
      "heldout_reasoning": {
        "passed": 108,
        "total": 129
      },
      "heldout_simulated_tools": {
        "passed": 201,
        "total": 201
      },
      "historical_regressions": {
        "passed": 11,
        "total": 16
      },
      "fresh_transfer": {
        "passed": 14,
        "total": 32
      },
      "all": {
        "passed": 343,
        "total": 436
      }
    },
    "upstream_base_comparison": {
      "model": "orcarouter/LFM3-1.7B-abliterated",
      "status": "not_evaluated",
      "scores": null
    },
    "paired_improvements_vs_modified_R3": 56,
    "paired_regressions_vs_modified_R3": 17
  },
  "weights": {
    "gguf_file": "LiquidVamp-R4-1.7B-BF16.gguf",
    "gguf_sha256": "55da5c3d28ddf7d34048347d0f3c797cc251e0c0dda9ae11b16d59e3cbfede41",
    "checkpoint": "Full merged model; supports further training without the original predecessor",
    "architecture": "lfm3",
    "quantization": "BF16 GGUF with inherited F32 tensors preserved"
  },
  "deployment": {
    "abliterated_by_lineage": true,
    "abliterated_upstream": "orcarouter/LFM3-1.7B-abliterated",
    "context_window_tokens": 131072,
    "context_window_scope": "Shared input, history, optional reasoning and output",
    "configured_output_budget_tokens": 16384,
    "architecture_max_position_setting": 262144,
    "backend": "llama.cpp b12179",
    "build": "669204ee2",
    "acceleration": "Vulkan/RADV",
    "os": "Ubuntu",
    "gpu": "AMD Radeon RX 6600",
    "gpu_memory_gb": 8,
    "ngram_compatible": true,
    "speculative_decoder": "ngram-mod",
    "ngram_match_tokens": 24,
    "ngram_min_proposal_tokens": 8,
    "ngram_max_proposal_tokens": 16,
    "kv_cache": "F16/F16",
    "parallel_slots": 1,
    "high_context_validation": "128K service initialization and basic API operation; scored and controlled speed tests at 8K",
    "full_128K_quality_benchmark": false
  }
}