Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75999.17790274702,
-      "required_pages_per_second_upper_estimate": 0.006587416122454718
+      "required_token_positions_per_second": 75936.52903197959,
+      "required_pages_per_second_upper_estimate": 0.006581985877118709
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9499.897237843377,
-      "required_pages_per_second_upper_estimate": 0.0037896510442968633
+      "required_token_positions_per_second": 9492.066128997449,
+      "required_pages_per_second_upper_estimate": 0.0037865270978927113
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18999.794475686755,
-      "required_pages_per_second_upper_estimate": 0.0043952680669880236
+      "required_token_positions_per_second": 18984.132257994897,
+      "required_pages_per_second_upper_estimate": 0.004391644888570629
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28499.691713530134,
-      "required_pages_per_second_upper_estimate": 0.0017354816990275677
+      "required_token_positions_per_second": 28476.198386992342,
+      "required_pages_per_second_upper_estimate": 0.0017340510787013745
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28499.691713530134,
-      "required_pages_per_second_upper_estimate": 0.002193985160303538
+      "required_token_positions_per_second": 28476.198386992342,
+      "required_pages_per_second_upper_estimate": 0.0021921765789929687
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28499.691713530134,
-      "required_pages_per_second_upper_estimate": 0.35387078874964467
+      "required_token_positions_per_second": 28476.198386992342,
+      "required_pages_per_second_upper_estimate": 0.3535790802611513
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1137852300705049,
-    "producer_p95_seconds": 0.3451631110001472,
-    "gpu_update_seconds": 0.9701157569670904,
-    "producer_batches_per_second": 8.788486865829324,
-    "gpu_batches_per_second": 1.0308048218146026,
-    "producer_headroom": 8.525849588439346,
+    "producer_batch_seconds": 0.4087256062186384,
+    "producer_p95_seconds": 0.35336289899714757,
+    "gpu_update_seconds": 0.9709161182354082,
+    "producer_batches_per_second": 2.4466291927525403,
+    "gpu_batches_per_second": 1.029955092122119,
+    "producer_headroom": 2.375471718588727,
     "sustainable": true,
-    "buffer_seconds": 60.147176931959606,
+    "buffer_seconds": 60.19679933059531,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291794.977825245,
+    "generation_bytes_per_second": 270864.88819898857,
     "upload_bytes_per_second": 5266697.155774625,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.612409601860165,
+    "estimated_full_shards_per_hour": 58.121299595615795,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -364,3 +364,4 @@
 {"step": 273500, "loss": 0.6166048962622881, "lr": 0.0003, "grad_norm": 0.1644891949652567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40031603668, "tokens_per_second": 148641.05553318723, "data_wait_s": 1.8188002286478877e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 273600, "loss": 0.5984455365687609, "lr": 0.0003, "grad_norm": 0.1406793601643985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40046282027, "tokens_per_second": 149426.92946310755, "data_wait_s": 9.884999599307775e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 273700, "loss": 0.5720657892525196, "lr": 0.0003, "grad_norm": 0.15198303023170256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40060943559, "tokens_per_second": 150887.69928129172, "data_wait_s": 9.119998139794916e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 273800, "loss": 0.6430566664785147, "lr": 0.0003, "grad_norm": 0.16033061241497748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40075615549, "tokens_per_second": 148455.10996529672, "data_wait_s": 1.4830002328380942e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a0b4210e422f47f5a2e0a20cf354656d157e9fc5842a91d65bd6def36ef55d3
-size 780834
+oid sha256:84b7abdf08797c4c2dfeaa11ac59dec189174b54f7bf19a9360199c563f9827a
+size 782952