{
  "model_details": {
    "provider": "e-infra",
    "model_requested": "deepseek-v4-pro-thinking",
    "model_for_requests": "deepseek-v4-pro-thinking",
    "api_base_url": "https://llm.ai.e-infra.cz/v1",
    "chat_completions_endpoint": "https://llm.ai.e-infra.cz/v1/chat/completions",
    "model_metadata": {
      "model_name": "deepseek-v4-pro-thinking",
      "litellm_model": "deepseek-v4-pro-thinking",
      "context_size": 1048576,
      "max_output_tokens": 64000,
      "model_source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro",
      "quantization": "fp4",
      "capabilities": [
        "chat",
        "tools"
      ],
      "extra_body": {
        "max_tokens": 64000,
        "max_completion_tokens": 64000
      },
      "input_cost_per_token": 0.00000174,
      "output_cost_per_token": 0.00000348
    },
    "model_source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro",
    "quantization": "fp4",
    "context_size": 1048576,
    "max_output_tokens": 64000,
    "input_cost_per_token": 0.00000174,
    "output_cost_per_token": 0.00000348
  },
  "run_config": {
    "input": [
      "data/input/prepositions.csv"
    ],
    "labels": "",
    "task_name": "Err. correct. prepositions",
    "task_description": "Correcting errors in prepositions in a transcription of learners' dialogue.",
    "tags": "error;correction;preposition;English",
    "model": "deepseek-v4-pro-thinking",
    "temperature": null,
    "top_p": null,
    "top_k": null,
    "service_tier": "standard",
    "verbosity": null,
    "reasoning_effort": null,
    "thinking_level": null,
    "effort": null,
    "strict_control_acceptance": true,
    "provider": "e-infra",
    "system_prompt": "You are a meticulous annotator focusing on grammatical errors in the use of prepositions.\nYou will receive a short excerpt of a transcribed dialogue where the node word is a preposition. \nDecide whether it is a correct preposition given the context. \nReturn the correct preposition (even if it already is correct). If the preposition should be omitted (i.e. the grammatical correct solution is leaving it out), return 0.",
    "system_prompt_b64": null,
    "few_shot_examples": 0,
    "prompt_layout": "standard",
    "cache_pad_target_tokens": 0,
    "prompt_cache_key": null,
    "gemini_cached_content": null,
    "requesty_auto_cache": null,
    "vertex_auto_adc_login": null,
    "vertex_access_token_refresh_seconds": null,
    "create_gemini_cache": false,
    "gemini_cache_ttl": 3600,
    "gemini_cache_ttl_autoupdate": true,
    "keep_gemini_cache": false,
    "enable_cot": true,
    "no_explanation": false,
    "logprobs": true,
    "calibration": true,
    "confusion_heatmap": true,
    "api_key_var": "E-INFRA_API_KEY",
    "api_base_var": "E-INFRA_BASE_URL",
    "max_retries": 3,
    "retry_delay": 5,
    "request_interval_ms": 0,
    "request_timeout_seconds": 30,
    "threads": 3,
    "prompt_log_detail": "full",
    "flush_rows": 100,
    "flush_seconds": 2,
    "validator_cmd": null,
    "validator_args": "",
    "validator_timeout": 5,
    "validator_prompt_max_candidates": 50,
    "validator_prompt_max_chars": 8000,
    "validator_exhausted_policy": "accept_blank_confidence",
    "validator_debug": false,
    "log_level": "INFO"
  },
  "source_input_csv": "data/input/prepositions.csv",
  "source_output_csv": "data/output/prepositions__einfra__deepseekv4prothinking__2026-05-08-22-34.csv",
  "source_labels_csv": "",
  "cache_padding": {
    "enabled": false,
    "target_shared_prefix_tokens": 0,
    "calibration_shared_prefix_tokens": null,
    "target_prompt_tokens": 0,
    "calibration_prompt_tokens": null,
    "calibration_example_id": null,
    "applied_padding_tokens_estimate": 0,
    "examples_with_padding_applied": 0
  },
  "request_control_summary": {
    "configured": {},
    "attempts_total": 1745,
    "attempts_with_control_telemetry": 1286,
    "per_control": {
      "reasoning_effort": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "thinking_level": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "effort": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "verbosity": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "prompt_cache_key": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "gemini_cached_content": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      },
      "requesty_auto_cache": {
        "configured_value": null,
        "requested_attempts": 0,
        "sent_attempts": 0,
        "accepted_attempts": 0,
        "rejected_attempts": 0,
        "missing_from_final_request_attempts": 0,
        "acceptance_rate": null,
        "rejected_reasons": {},
        "rejected_example_ids": []
      }
    }
  },
  "usage_metadata_summary": {
    "attempts_total": 1745,
    "attempts_with_usage_metadata": 1286,
    "attempts_with_cached_token_signals": 0,
    "cached_tokens_total_estimate": 0,
    "cache_read_tokens_total": 0,
    "cache_write_tokens_total": 0,
    "cache_token_fields_totals": {},
    "attempts_with_gemini_cached_content_token_signals": 0,
    "gemini_cached_content_token_count_total": 0,
    "gemini_cached_content_token_fields_totals": {}
  },
  "token_usage_totals": {
    "attempts_total": 1745,
    "attempts_with_token_usage": 1286,
    "attempts_with_output_tokens": 1286,
    "attempts_with_cached_input_tokens": 0,
    "attempts_with_thinking_tokens": 0,
    "input_tokens_total": 759237,
    "cached_input_tokens_total": 0,
    "non_cached_input_tokens_total": 759237,
    "output_tokens_total": 624971,
    "thinking_tokens_total": 0,
    "output_tokens_definition": "total_tokens - prompt_tokens (or completion_tokens + thinking_tokens fallback)"
  },
  "truth_label_count": 1283,
  "prediction_count": 1283,
  "evaluated_example_count": 1283,
  "calibration_metrics": {
    "available": true,
    "sample_count": 1283,
    "bin_count": 10,
    "ece": 0.08475448168356702,
    "mce": 0.4285714285714287,
    "brier_score": 0.11039828526890112
  },
  "first_prompt_timestamp": "2026-05-08T20:34:03.916617Z",
  "last_prompt_timestamp": "2026-05-08T22:49:50.857054Z",
  "overall_time_seconds": 8146.940437,
  "overall_time_human": "2h 15m 47s",
  "accuracy": 0.8737334372564303,
  "cohen_kappa": 0.7405748278474599,
  "macro_precision": 0.38740276460350387,
  "macro_recall": 0.3827089959995269,
  "macro_f1": 0.3698434033822943,
  "per_label": {
    "0": {
      "precision": 0.20454545454545456,
      "recall": 0.72,
      "f1": 0.3185840707964602,
      "support": 25
    },
    "CORRECT": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "about": {
      "precision": 0.5,
      "recall": 0.5,
      "f1": 0.5,
      "support": 2
    },
    "around": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 1
    },
    "as": {
      "precision": 1,
      "recall": 1,
      "f1": 1,
      "support": 1
    },
    "at": {
      "precision": 0.9187817258883249,
      "recall": 0.900497512437811,
      "f1": 0.9095477386934674,
      "support": 201
    },
    "behind": {
      "precision": 1,
      "recall": 1,
      "f1": 1,
      "support": 1
    },
    "correct": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "during": {
      "precision": 0.6666666666666666,
      "recall": 0.5,
      "f1": 0.5714285714285715,
      "support": 4
    },
    "extra_preposition": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "from": {
      "precision": 0.3333333333333333,
      "recall": 0.5,
      "f1": 0.4,
      "support": 2
    },
    "in": {
      "precision": 0.9525462962962963,
      "recall": 0.8955386289445049,
      "f1": 0.9231632080762758,
      "support": 919
    },
    "incorrect": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "into": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 1
    },
    "of": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 2
    },
    "on": {
      "precision": 0.8617021276595744,
      "recall": 0.8804347826086957,
      "f1": 0.870967741935484,
      "support": 92
    },
    "since": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    },
    "to": {
      "precision": 0.9230769230769231,
      "recall": 0.375,
      "f1": 0.5333333333333333,
      "support": 32
    },
    "with": {
      "precision": 0,
      "recall": 0,
      "f1": 0,
      "support": 0
    }
  },
  "labels": [
    "0",
    "CORRECT",
    "about",
    "around",
    "as",
    "at",
    "behind",
    "correct",
    "during",
    "extra_preposition",
    "from",
    "in",
    "incorrect",
    "into",
    "of",
    "on",
    "since",
    "to",
    "with"
  ],
  "label_count": 19,
  "total_examples": 1283,
  "confusion_matrix_sparse": [
    [
      0,
      0,
      18
    ],
    [
      0,
      5,
      1
    ],
    [
      0,
      8,
      1
    ],
    [
      0,
      11,
      5
    ],
    [
      2,
      2,
      1
    ],
    [
      2,
      11,
      1
    ],
    [
      3,
      11,
      1
    ],
    [
      4,
      4,
      1
    ],
    [
      5,
      0,
      5
    ],
    [
      5,
      5,
      181
    ],
    [
      5,
      11,
      11
    ],
    [
      5,
      15,
      4
    ],
    [
      6,
      6,
      1
    ],
    [
      8,
      8,
      2
    ],
    [
      8,
      11,
      1
    ],
    [
      8,
      15,
      1
    ],
    [
      10,
      0,
      1
    ],
    [
      10,
      10,
      1
    ],
    [
      11,
      0,
      58
    ],
    [
      11,
      1,
      1
    ],
    [
      11,
      2,
      1
    ],
    [
      11,
      5,
      14
    ],
    [
      11,
      7,
      3
    ],
    [
      11,
      9,
      1
    ],
    [
      11,
      10,
      2
    ],
    [
      11,
      11,
      823
    ],
    [
      11,
      12,
      1
    ],
    [
      11,
      13,
      4
    ],
    [
      11,
      14,
      2
    ],
    [
      11,
      15,
      7
    ],
    [
      11,
      17,
      1
    ],
    [
      11,
      18,
      1
    ],
    [
      13,
      11,
      1
    ],
    [
      14,
      11,
      1
    ],
    [
      14,
      15,
      1
    ],
    [
      15,
      0,
      5
    ],
    [
      15,
      11,
      4
    ],
    [
      15,
      13,
      1
    ],
    [
      15,
      15,
      81
    ],
    [
      15,
      16,
      1
    ],
    [
      17,
      0,
      1
    ],
    [
      17,
      5,
      1
    ],
    [
      17,
      11,
      16
    ],
    [
      17,
      17,
      12
    ],
    [
      17,
      18,
      2
    ]
  ],
  "confusion_matrix_format": "sparse_triplets",
  "confusion_matrix_nonzero_cells": 45,
  "label_metrics_available": true,
  "estimated_cost_usd": 3.49597146,
  "pricing": {
    "status": "priced",
    "status_label": "priced",
    "provider_key": "e-infra",
    "matched_key": "deepseek-v4-pro-thinking",
    "resolved_key": "deepseek-v4-pro-thinking",
    "pricing_tier": "standard",
    "estimated_cost_usd": 3.49597146,
    "catalog_updated_at": "2026-05-27T15:36:49Z",
    "missing_buckets": []
  }
}