@tokcalc/mcp-server 0.1.3 → 0.2.0-alpha.1

package/src/lib/benchmark-parser-tokcalc.ts

@@ -1,247 +0,0 @@
-/**
- * Community benchmark submission template.
- *
- * Users download this JSON file, fill in their benchmark results,
- * and paste it into tokcalc's BenchmarkImport textarea.
- *
- * The tokcalc-standard parser recognizes this format and creates
- * a BenchmarkRecord with confidence_tier = "community_reproducible"
- * (if all required fields are filled) or "community_unverified" (if some are missing).
- *
- * This is the "tier 4" ingestion path from the Perplexity research:
- * "A tokcalc-standard JSON schema for community uploads."
- */
-
-export const BENCHMARK_TEMPLATE = `{
-  "format": "tokcalc-standard",
-  "version": "1.0",
-
-  "source": {
-    "submitter": "your-github-username",
-    "submit_date": "2026-09-24",
-    "raw_artifact_url": "https://github.com/your-username/your-repo/blob/main/benchmark-results.json",
-    "notes": "Describe how you ran this benchmark. Include the exact command you used."
-  },
-
-  "model": {
-    "model_id": "meta-llama/Llama-3.3-70B-Instruct",
-    "model_revision": "commit hash or release date",
-    "quantization_format": "FP8",
-    "weight_dtype": "float8_e4m3",
-    "kv_cache_dtype": "float16"
-  },
-
-  "hardware": {
-    "accelerator_vendor": "NVIDIA",
-    "accelerator_model": "H100 SXM5 80GB",
-    "gpu_count": 1,
-    "memory_per_gpu_gb": 80,
-    "interconnect": "NVLink",
-    "cloud_provider": "RunPod",
-    "instance_type": "H100 80GB SXM",
-    "region": "us-east-1",
-    "driver_version": "550.54.15",
-    "cuda_version": "12.4"
-  },
-
-  "engine": {
-    "serving_engine": "vLLM",
-    "engine_version": "0.29.0",
-    "tensor_parallel_size": 1,
-    "max_model_len": 32768,
-    "max_num_seqs": 256,
-    "max_num_batched_tokens": 8192,
-    "prefix_cache_enabled": true,
-    "chunked_prefill_enabled": true,
-    "attention_backend": "FlashInfer",
-    "command": "vllm serve meta-llama/Llama-3.3-70B-Instruct --quantization fp8 --tensor-parallel-size 1 --max-model-len 32768 --max-num-seqs 256"
-  },
-
-  "workload": {
-    "workload_type": "online_poisson",
-    "dataset_name": "ShareGPT",
-    "request_count": 1000,
-    "concurrency": 64,
-    "input_tokens_mean": 1024,
-    "input_tokens_p95": 4096,
-    "output_tokens_mean": 256,
-    "output_tokens_p95": 1024,
-    "streaming_enabled": true,
-    "sampling_config": {
-      "temperature": 0.7,
-      "top_p": 0.95,
-      "max_tokens": 512
-    }
-  },
-
-  "metrics": {
-    "successful_requests": 1000,
-    "failed_requests": 0,
-    "output_token_throughput_tps": 2450.5,
-    "total_token_throughput_tps": 11000.0,
-    "request_throughput_rps": 3.91,
-    "ttft_mean_ms": 112.3,
-    "ttft_p50_ms": 108.5,
-    "ttft_p95_ms": 145.2,
-    "ttft_p99_ms": 189.7,
-    "itl_mean_ms": 5.67,
-    "itl_p50_ms": 5.45,
-    "itl_p95_ms": 7.23,
-    "itl_p99_ms": 8.91,
-    "gpu_memory_peak_gb": 78.5,
-    "gpu_utilization_mean_pct": 92.5
-  }
-}`;
-
-/**
- * Parse a tokcalc-standard benchmark JSON.
- * This is the community submission format — users fill in the template
- * and paste it into the BenchmarkImport textarea.
- */
-import type {
-  BenchmarkParser,
-  BenchmarkRecord,
-  ConfidenceTier,
-} from "./benchmark-schema";
-import { generateBenchmarkId } from "./benchmark-schema";
-
-export const tokcalcStandardParser: BenchmarkParser = {
-  name: "tokcalc-standard",
-  version: "v1",
-  formatLabel: "tokcalc community benchmark",
-
-  detect(rawText: string): boolean {
-    try {
-      const data = JSON.parse(rawText);
-      return data?.format === "tokcalc-standard" || !!data?.metrics?.output_token_throughput_tps;
-    } catch {
-      return false;
-    }
-  },
-
-  parse(rawText: string): BenchmarkRecord | null {
-    try {
-      const data = JSON.parse(rawText);
-      if (!data) return null;
-
-      const model = data.model || {};
-      const hardware = data.hardware || {};
-      const engine = data.engine || {};
-      const workload = data.workload || {};
-      const metrics = data.metrics || {};
-      const source = data.source || {};
-
-      const modelId = model.model_id || "unknown";
-      const modelDisplayName = modelId.split("/").pop() || modelId;
-
-      // Determine confidence tier based on completeness
-      const hasFullConfig =
-        modelId !== "unknown" &&
-        hardware.accelerator_model &&
-        engine.serving_engine &&
-        metrics.output_token_throughput_tps > 0 &&
-        workload.request_count > 0;
-      const confidenceTier: ConfidenceTier = hasFullConfig
-        ? "community_reproducible"
-        : "community_unverified";
-
-      return {
-        benchmark_id: generateBenchmarkId(),
-        source_name: "community",
-        source_url: source.raw_artifact_url || "",
-        source_license: "CC-BY-SA-4.0",
-        ingestion_method: "uploaded_raw_artifact",
-        retrieved_at: new Date().toISOString(),
-        benchmark_started_at: source.submit_date,
-        raw_artifact_url: source.raw_artifact_url,
-        parser_name: "tokcalc-standard",
-        parser_version: "v1",
-        confidence_tier: confidenceTier,
-        verification_status: "schema_valid",
-
-        model_id: modelId,
-        model_display_name: modelDisplayName,
-        model_revision: model.model_revision || "unspecified",
-        architecture: model.architecture,
-        parameter_count_b: model.parameter_count_b,
-        context_window_tokens: engine.max_model_len,
-        quantization_format: model.quantization_format || "unspecified",
-        weight_dtype: model.weight_dtype,
-        kv_cache_dtype: model.kv_cache_dtype || "float16",
-
-        accelerator_vendor: hardware.accelerator_vendor || "NVIDIA",
-        accelerator_model: hardware.accelerator_model || "unspecified",
-        gpu_count: hardware.gpu_count || engine.tensor_parallel_size || 1,
-        memory_per_gpu_gb: hardware.memory_per_gpu_gb || 0,
-        host_cpu: hardware.host_cpu,
-        host_memory_gb: hardware.host_memory_gb,
-        interconnect: hardware.interconnect,
-        topology: hardware.topology,
-        cloud_provider: hardware.cloud_provider,
-        instance_type: hardware.instance_type,
-        region: hardware.region,
-        driver_version: hardware.driver_version,
-        cuda_rocm_metal_version: hardware.cuda_version,
-
-        serving_engine: engine.serving_engine || "unspecified",
-        engine_version: engine.engine_version || "unspecified",
-        engine_command: engine.command,
-        engine_config: {
-          tensor_parallel_size: engine.tensor_parallel_size,
-          max_model_len: engine.max_model_len,
-          max_num_seqs: engine.max_num_seqs,
-          max_num_batched_tokens: engine.max_num_batched_tokens,
-          prefix_cache_enabled: engine.prefix_cache_enabled,
-          chunked_prefill_enabled: engine.chunked_prefill_enabled,
-          attention_backend: engine.attention_backend,
-        },
-
-        workload_type: workload.workload_type || "online_poisson",
-        dataset_name: workload.dataset_name,
-        request_count: workload.request_count || 0,
-        concurrency: workload.concurrency || 1,
-        arrival_rate_rps: workload.arrival_rate_rps,
-        input_tokens_mean: workload.input_tokens_mean || 0,
-        input_tokens_p50: workload.input_tokens_p50,
-        input_tokens_p95: workload.input_tokens_p95,
-        input_tokens_p99: workload.input_tokens_p99,
-        output_tokens_mean: workload.output_tokens_mean || 0,
-        output_tokens_p50: workload.output_tokens_p50,
-        output_tokens_p95: workload.output_tokens_p95,
-        output_tokens_p99: workload.output_tokens_p99,
-        shared_prefix_tokens_mean: workload.shared_prefix_tokens_mean,
-        shared_prefix_hit_rate: workload.shared_prefix_hit_rate,
-        streaming_enabled: workload.streaming_enabled ?? true,
-        sampling_config: workload.sampling_config,
-
-        successful_requests: metrics.successful_requests ?? workload.request_count ?? 0,
-        failed_requests: metrics.failed_requests ?? 0,
-        request_throughput_rps: metrics.request_throughput_rps,
-        output_token_throughput_tps: metrics.output_token_throughput_tps || 0,
-        total_token_throughput_tps: metrics.total_token_throughput_tps,
-        ttft_mean_ms: metrics.ttft_mean_ms,
-        ttft_p50_ms: metrics.ttft_p50,
-        ttft_p95_ms: metrics.ttft_p95_ms,
-        ttft_p99_ms: metrics.ttft_p99_ms,
-        itl_mean_ms: metrics.itl_mean_ms,
-        itl_p50_ms: metrics.itl_p50,
-        itl_p95_ms: metrics.itl_p95_ms,
-        itl_p99_ms: metrics.itl_p99_ms,
-        e2e_latency_p50_ms: metrics.e2e_latency_p50_ms,
-        e2e_latency_p95_ms: metrics.e2e_latency_p95_ms,
-        e2e_latency_p99_ms: metrics.e2e_latency_p99_ms,
-        gpu_memory_peak_gb: metrics.gpu_memory_peak_gb,
-        gpu_utilization_mean_pct: metrics.gpu_utilization_mean_pct,
-        power_mean_w: metrics.power_mean_w,
-        energy_per_output_token_j: metrics.energy_per_output_token_j,
-
-        metric_definition_version: "tokcalc-v1",
-        citation_text: `Community benchmark: ${metrics.output_token_throughput_tps || 0} tok/s, ${workload.request_count || 0} requests, submitted by ${source.submitter || "anonymous"}`,
-        attribution_required: true,
-        commercial_redistribution_allowed: "yes",
-      };
-    } catch {
-      return null;
-    }
-  },
-};

package/src/lib/benchmark-parser-trtllm.ts

@@ -1,152 +0,0 @@
-/**
- * TensorRT-LLM benchmark parser.
- *
- * TRT-LLM's `benchmark_serving.py` produces JSON output with fields like:
- *   - throughput (tokens/s)
- *   - latency_avg (ms) — end-to-end latency
- *   - first_token_latency_avg (ms) — TTFT
- *   - inter_token_latency_avg (ms) — ITL
- *   - total_input_tokens, total_output_tokens
- *   - num_requests
- *   - model
- *
- * Refs: https://nvidia.github.io/TensorRT-LLM/1.3.0rc4/developer-guide/perf-benchmarking.html
- */
-
-import type {
-  BenchmarkParser,
-  BenchmarkRecord,
-  ConfidenceTier,
-} from "./benchmark-schema";
-import { generateBenchmarkId } from "./benchmark-schema";
-
-export const trtllmParser: BenchmarkParser = {
-  name: "trtllm-json",
-  version: "v1",
-  formatLabel: "TensorRT-LLM benchmark JSON",
-
-  detect(rawText: string): boolean {
-    try {
-      const data = JSON.parse(rawText);
-      return !!(
-        data &&
-        (data.throughput ||
-          data.first_token_latency_avg ||
-          data.inter_token_latency_avg ||
-          data.latency_avg ||
-          data.total_input_tokens ||
-          data.total_output_tokens)
-      );
-    } catch {
-      return false;
-    }
-  },
-
-  parse(rawText: string): BenchmarkRecord | null {
-    try {
-      const data = JSON.parse(rawText);
-      if (!data) return null;
-
-      const modelId = data.model || data.model_id || data.model_name || "unknown";
-      const modelDisplayName = modelId.split("/").pop() || modelId;
-
-      const outputThroughput = data.throughput ?? data.output_throughput ?? 0;
-      const meanTtft =
-        data.first_token_latency_avg ?? data.ttft_mean ?? data.mean_ttft_ms;
-      const meanItl =
-        data.inter_token_latency_avg ?? data.itl_mean ?? data.mean_itl_ms;
-      const meanE2E = data.latency_avg ?? data.e2e_latency_mean;
-
-      const numRequests = data.num_requests ?? data.completed ?? 0;
-      const inputLen = Number(
-        data.input_len ?? data.mean_input_len ?? data.total_input_tokens / Math.max(numRequests, 1) ?? 0,
-      );
-      const outputLen = Number(
-        data.output_len ?? data.mean_output_len ?? data.total_output_tokens / Math.max(numRequests, 1) ?? 0,
-      );
-      const concurrency = data.concurrency ?? data.num_procs ?? 1;
-
-      const confidenceTier: ConfidenceTier =
-        modelId !== "unknown" && outputThroughput > 0 && numRequests > 0
-          ? "community_reproducible"
-          : "community_unverified";
-
-      return {
-        benchmark_id: generateBenchmarkId(),
-        source_name: "tensorrt_llm",
-        source_url: data.source_url || "https://nvidia.github.io/TensorRT-LLM/",
-        source_license: "Apache-2.0",
-        ingestion_method: "uploaded_raw_artifact",
-        retrieved_at: new Date().toISOString(),
-        parser_name: "trtllm-json",
-        parser_version: "v1",
-        confidence_tier: confidenceTier,
-        verification_status: "schema_valid",
-
-        model_id: modelId,
-        model_display_name: modelDisplayName,
-        model_revision: data.model_revision || "unspecified",
-        architecture: data.architecture,
-        parameter_count_b: data.parameter_count_b,
-        context_window_tokens: data.max_model_len,
-        quantization_format: data.quantization || data.dtype || "unspecified",
-        weight_dtype: data.dtype,
-        kv_cache_dtype: data.kv_cache_dtype || "float16",
-
-        accelerator_vendor: "NVIDIA",
-        accelerator_model: data.gpu_model || "unspecified",
-        gpu_count: data.tensor_parallel_size || data.gpu_count || 1,
-        memory_per_gpu_gb: data.gpu_memory || 0,
-        driver_version: data.driver_version,
-        cuda_rocm_metal_version: data.cuda_version,
-
-        serving_engine: "TensorRT-LLM",
-        engine_version: data.trtllm_version || "unspecified",
-        engine_command: data.command,
-        engine_config: {
-          tensor_parallel_size: data.tensor_parallel_size,
-          max_model_len: data.max_model_len,
-          max_num_seqs: data.max_num_seqs,
-          max_num_batched_tokens: data.max_num_batched_tokens,
-          prefix_cache_enabled: data.enable_prefix_caching,
-          chunked_prefill_enabled: data.enable_chunked_prefill,
-          attention_backend: "TensorRT-LLM",
-        },
-
-        workload_type: data.workload_type || "online_poisson",
-        dataset_name: data.dataset_name || data.dataset,
-        request_count: numRequests,
-        concurrency,
-        input_tokens_mean: inputLen,
-        output_tokens_mean: outputLen,
-        streaming_enabled: data.streaming ?? true,
-
-        successful_requests: numRequests,
-        failed_requests: data.failed ?? 0,
-        output_token_throughput_tps: outputThroughput,
-        total_token_throughput_tps: data.total_throughput,
-        request_throughput_rps: data.request_throughput,
-        ttft_mean_ms: meanTtft,
-        ttft_p50_ms: data.first_token_latency_p50,
-        ttft_p95_ms: data.first_token_latency_p95,
-        ttft_p99_ms: data.first_token_latency_p99,
-        itl_mean_ms: meanItl,
-        itl_p50_ms: data.inter_token_latency_p50,
-        itl_p95_ms: data.inter_token_latency_p95,
-        itl_p99_ms: data.inter_token_latency_p99,
-        e2e_latency_p50_ms: data.latency_p50,
-        e2e_latency_p95_ms: data.latency_p95,
-        e2e_latency_p99_ms: data.latency_p99,
-
-        gpu_memory_peak_gb: data.peak_gpu_memory,
-
-        metric_definition_version: "trtllm-v1",
-        citation_text: `TRT-LLM benchmark: ${outputThroughput.toFixed(2)} tok/s, TTFT mean ${meanTtft?.toFixed(1) || "?"} ms, ${numRequests} requests`,
-        attribution_required: true,
-        commercial_redistribution_allowed: "yes",
-      };
-    } catch {
-      return null;
-    }
-  },
-};

package/src/lib/benchmark-parser-vllm.ts

@@ -1,198 +0,0 @@
-/**
- * vLLM benchmark JSON parser.
- *
- * vLLM's `benchmark_serving.py` produces JSON output with fields like:
- *   - output_throughput: output tokens/s
- *   - mean_ttft_ms: time to first token
- *   - p99_ttft_ms
- *   - mean_itl_ms: inter-token latency
- *   - mean_tpot_ms: time per output token
- *   - total_token_throughput
- *   - duration
- *   - completed / total_input_tokens / total_output_tokens
- *
- * Example vLLM benchmark output:
- * {
- *   "elapsed_time": 100.5,
- *   "num_input_tokens": 1024000,
- *   "num_output_tokens": 256000,
- *   "request_throughput": 1.73,
- *   "output_throughput": 382.89,
- *   "total_token_throughput": 619.85,
- *   "mean_ttft_ms": 71.54,
- *   "median_ttft_ms": 70.21,
- *   "p99_ttft_ms": 79.49,
- *   "mean_tpot_ms": 12.45,
- *   "median_tpot_ms": 12.30,
- *   "p99_tpot_ms": 14.87,
- *   "mean_itl_ms": 7.74,
- *   "median_itl_ms": 7.68,
- *   "p99_itl_ms": 8.92,
- *   "input_len": "1024",
- *   "output_len": "256",
- *   "num_requests": 1000,
- *   "model_id": "meta-llama/Llama-3.1-70B-Instruct",
- *   ...
- * }
- *
- * The actual field names vary by vLLM version. This parser handles the
- * common field aliases across v0.5.x → v0.29.x.
- */
-
-import type {
-  BenchmarkParser,
-  BenchmarkRecord,
-  ConfidenceTier,
-} from "./benchmark-schema";
-import { generateBenchmarkId } from "./benchmark-schema";
-
-export const vllmParser: BenchmarkParser = {
-  name: "vllm-json",
-  version: "v1",
-  formatLabel: "vLLM benchmark JSON",
-
-  detect(rawText: string): boolean {
-    try {
-      const data = JSON.parse(rawText);
-      // Check for vLLM-specific fields. Require at least 2 to avoid
-      // catching TRT-LLM or SGLang output which may share `num_requests`.
-      const vllmFields = [
-        data.output_throughput,
-        data.total_token_throughput,
-        data.mean_ttft_ms,
-        data.mean_itl_ms,
-        data.mean_tpot_ms,
-        data.request_throughput,
-      ].filter((v) => v !== undefined && v !== null);
-      return vllmFields.length >= 2;
-    } catch {
-      return false;
-    }
-  },
-
-  parse(rawText: string): BenchmarkRecord | null {
-    try {
-      const data = JSON.parse(rawText);
-      if (!data) return null;
-
-      // Extract model info
-      const modelId = data.model_id || data.model || "unknown";
-      const modelDisplayName = modelId.split("/").pop() || modelId;
-
-      // Extract throughput metrics (handle field name variations)
-      const outputThroughput = data.output_throughput ?? data.output_token_throughput ?? 0;
-      const totalThroughput = data.total_token_throughput ?? 0;
-      const requestThroughput = data.request_throughput ?? 0;
-
-      // Extract latency metrics
-      const meanTtft = data.mean_ttft_ms ?? data.mean_ttft;
-      const p50Ttft = data.median_ttft_ms ?? data.p50_ttft_ms ?? data.ttft_p50;
-      const p95Ttft = data.p95_ttft_ms ?? data.ttft_p95;
-      const p99Ttft = data.p99_ttft_ms ?? data.p99_ttft;
-
-      const meanItl = data.mean_itl_ms ?? data.mean_itl;
-      const p50Itl = data.median_itl_ms ?? data.p50_itl_ms ?? data.itl_p50;
-      const p95Itl = data.p95_itl_ms ?? data.itl_p95;
-      const p99Itl = data.p99_itl_ms ?? data.p99_itl;
-
-      const meanTpot = data.mean_tpot_ms ?? data.mean_tpot;
-      const p99Tpot = data.p99_tpot_ms ?? data.p99_tpot;
-
-      // Extract workload info
-      const numRequests = data.num_requests ?? data.completed ?? 0;
-      const inputLen = Number(data.input_len ?? data.mean_input_len ?? data.input_tokens_mean ?? 0);
-      const outputLen = Number(data.output_len ?? data.mean_output_len ?? data.output_tokens_mean ?? 0);
-      const concurrency = data.num_procs ?? data.concurrency ?? 1;
-
-      // Extract duration
-      const elapsed = data.elapsed_time ?? data.duration ?? 0;
-
-      // Determine confidence tier
-      const hasFullConfig = !!data.model_id && !!data.output_throughput && numRequests > 0;
-      const confidenceTier: ConfidenceTier = hasFullConfig
-        ? "community_reproducible"
-        : "community_unverified";
-
-      const record: BenchmarkRecord = {
-        benchmark_id: generateBenchmarkId(),
-        source_name: "vllm",
-        source_url: data.source_url || "https://docs.vllm.ai/en/latest/benchmarking/cli/",
-        source_license: "Apache-2.0",
-        ingestion_method: "uploaded_raw_artifact",
-        retrieved_at: new Date().toISOString(),
-        benchmark_started_at: data.timestamp || data.run_time,
-        parser_name: "vllm-json",
-        parser_version: "v1",
-        confidence_tier: confidenceTier,
-        verification_status: "schema_valid",
-
-        model_id: modelId,
-        model_display_name: modelDisplayName,
-        model_revision: data.model_revision || data.commit || "unspecified",
-        architecture: data.architecture || "dense",
-        parameter_count_b: data.parameter_count_b,
-        context_window_tokens: data.max_model_len,
-        quantization_format: data.quantization || data.dtype || "unspecified",
-        weight_dtype: data.dtype || data.weight_dtype,
-        kv_cache_dtype: data.kv_cache_dtype || "float16",
-
-        accelerator_vendor: data.gpu_vendor || "NVIDIA",
-        accelerator_model: data.gpu_model || "unspecified",
-        gpu_count: data.tensor_parallel_size || data.gpu_count || 1,
-        memory_per_gpu_gb: data.gpu_memory || 0,
-        driver_version: data.driver_version,
-        cuda_rocm_metal_version: data.cuda_version,
-
-        serving_engine: "vLLM",
-        engine_version: data.vllm_version || "unspecified",
-        engine_command: data.command,
-        engine_config: {
-          tensor_parallel_size: data.tensor_parallel_size,
-          max_model_len: data.max_model_len,
-          max_num_seqs: data.max_num_seqs,
-          max_num_batched_tokens: data.max_num_batched_tokens,
-          prefix_cache_enabled: data.enable_prefix_caching ?? data.prefix_cache_enabled,
-          chunked_prefill_enabled: data.enable_chunked_prefill,
-          attention_backend: data.attention_backend,
-        },
-
-        workload_type: data.workload_type || "online_poisson",
-        dataset_name: data.dataset_name || data.dataset,
-        request_count: numRequests,
-        concurrency,
-        input_tokens_mean: inputLen,
-        output_tokens_mean: outputLen,
-        streaming_enabled: data.streaming ?? true,
-
-        successful_requests: numRequests,
-        failed_requests: data.failed ?? (data.total_requests ? data.total_requests - numRequests : 0),
-        request_throughput_rps: requestThroughput,
-        output_token_throughput_tps: outputThroughput,
-        total_token_throughput_tps: totalThroughput,
-        ttft_mean_ms: meanTtft,
-        ttft_p50_ms: p50Ttft,
-        ttft_p95_ms: p95Ttft,
-        ttft_p99_ms: p99Ttft,
-        itl_mean_ms: meanItl,
-        itl_p50_ms: p50Itl,
-        itl_p95_ms: p95Itl,
-        itl_p99_ms: p99Itl,
-        tpot_mean_ms: meanTpot,
-        e2e_latency_p99_ms: data.mean_lantecy_ms || data.e2e_latency_p99_ms,
-
-        gpu_memory_peak_gb: data.peak_gpu_memory || data.gpu_memory_peak_gb,
-        gpu_utilization_mean_pct: data.gpu_utilization,
-
-        metric_definition_version: "vllm-v1",
-        citation_text: `vLLM benchmark: ${outputThroughput.toFixed(2)} output tok/s, TTFT mean ${meanTtft?.toFixed(1) || "?"} ms, ${numRequests} requests`,
-        attribution_required: true,
-        commercial_redistribution_allowed: "yes",
-      };
-
-      return record;
-    } catch (e) {
-      console.error("vLLM parse error:", e);
-      return null;
-    }
-  },
-};