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": 75594.51657615908,
-      "required_pages_per_second_upper_estimate": 0.00655234110427083
+      "required_token_positions_per_second": 75310.7424897172,
+      "required_pages_per_second_upper_estimate": 0.006527744285676896
     },
     {
       "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": 9449.314572019885,
-      "required_pages_per_second_upper_estimate": 0.003769472862621623
+      "required_token_positions_per_second": 9413.84281121465,
+      "required_pages_per_second_upper_estimate": 0.0037553226468863294
     },
     {
       "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": 18898.62914403977,
-      "required_pages_per_second_upper_estimate": 0.00437186524796052
+      "required_token_positions_per_second": 18827.6856224293,
+      "required_pages_per_second_upper_estimate": 0.004355453712799275
     },
     {
       "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": 28347.943716059654,
-      "required_pages_per_second_upper_estimate": 0.0017262410421418277
+      "required_token_positions_per_second": 28241.528433643947,
+      "required_pages_per_second_upper_estimate": 0.0017197609097605489
     },
     {
       "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": 28347.943716059654,
-      "required_pages_per_second_upper_estimate": 0.002182303179392918
+      "required_token_positions_per_second": 28241.528433643947,
+      "required_pages_per_second_upper_estimate": 0.0021741110363762014
     },
     {
       "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": 28347.943716059654,
-      "required_pages_per_second_upper_estimate": 0.35198658648893866
+      "required_token_positions_per_second": 28241.528433643947,
+      "required_pages_per_second_upper_estimate": 0.35066526483037547
     }
   ],
   "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": 0.2576498914058334,
-    "producer_p95_seconds": 0.41716916199948173,
-    "gpu_update_seconds": 0.975308836398489,
-    "producer_batches_per_second": 3.8812358683468835,
-    "gpu_batches_per_second": 1.025316251304241,
-    "producer_headroom": 3.785403638545478,
+    "producer_batch_seconds": 0.4581010935467589,
+    "producer_p95_seconds": 0.40679565299797105,
+    "gpu_update_seconds": 0.978983841648709,
+    "producer_batches_per_second": 2.18292428044145,
+    "gpu_batches_per_second": 1.0214673189252006,
+    "producer_headroom": 2.1370475980948145,
     "sustainable": true,
-    "buffer_seconds": 60.469147856706314,
+    "buffer_seconds": 60.69699818221996,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445589.72292000084,
+    "generation_bytes_per_second": 431117.15599402675,
     "upload_bytes_per_second": 4022126.850080259,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.61318174076098,
+    "estimated_full_shards_per_hour": 92.50770578256227,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -354,3 +354,4 @@
 {"step": 350400, "loss": 0.5734471511095762, "lr": 0.0003, "grad_norm": 0.1576187271171644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51289254835, "tokens_per_second": 147763.8878159974, "data_wait_s": 1.3365002814680338e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350500, "loss": 0.5647007338702679, "lr": 0.0003, "grad_norm": 0.16194227915709652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51303831394, "tokens_per_second": 148074.64501590794, "data_wait_s": 1.7079997633118182e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350600, "loss": 0.588532218709588, "lr": 0.0003, "grad_norm": 0.16141536980723187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51318537286, "tokens_per_second": 148859.55574717984, "data_wait_s": 1.3267999747768044e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 350700, "loss": 0.5680049220100045, "lr": 0.0003, "grad_norm": 0.15940457579135126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51333187567, "tokens_per_second": 149293.90727806013, "data_wait_s": 1.622399577172473e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27cbcfc157d4893512abc8447921959a84c8fc7b5bf0d49d21feed5d1b0b5338
-size 759654
+oid sha256:d5ba57be5244a10a0b382412bb23c0dea4cb80d07f09fc772c151f6981bbe6f8
+size 761772