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": 76068.7019570255,
-      "required_pages_per_second_upper_estimate": 0.006593442291272473
+      "required_token_positions_per_second": 75934.55418140335,
+      "required_pages_per_second_upper_estimate": 0.006581814702075973
     },
     {
       "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": 9508.587744628188,
-      "required_pages_per_second_upper_estimate": 0.00379311781738798
+      "required_token_positions_per_second": 9491.819272675419,
+      "required_pages_per_second_upper_estimate": 0.0037864286232150227
     },
     {
       "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": 19017.175489256377,
-      "required_pages_per_second_upper_estimate": 0.004399288858581964
+      "required_token_positions_per_second": 18983.638545350837,
+      "required_pages_per_second_upper_estimate": 0.004391530676839471
     },
     {
       "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": 28525.76323388457,
-      "required_pages_per_second_upper_estimate": 0.0017370693178304568
+      "required_token_positions_per_second": 28475.457818026258,
+      "required_pages_per_second_upper_estimate": 0.001734005981936805
     },
     {
       "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": 28525.76323388457,
-      "required_pages_per_second_upper_estimate": 0.0021959922181110096
+      "required_token_positions_per_second": 28475.457818026258,
+      "required_pages_per_second_upper_estimate": 0.0021921195679439334
     },
     {
       "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": 28525.76323388457,
-      "required_pages_per_second_upper_estimate": 0.35419450977668115
+      "required_token_positions_per_second": 28475.457818026258,
+      "required_pages_per_second_upper_estimate": 0.3535698848731174
     }
   ],
   "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": 0.08551762317966904,
-    "producer_p95_seconds": 0.21395238999866706,
-    "gpu_update_seconds": 0.9692291061000636,
-    "producer_batches_per_second": 11.6934961803024,
-    "gpu_batches_per_second": 1.0317478021514963,
-    "producer_headroom": 11.333676850019005,
+    "producer_batch_seconds": 0.07997379347652611,
+    "producer_p95_seconds": 0.19506554100007634,
+    "gpu_update_seconds": 0.9709413691146191,
+    "producer_batches_per_second": 12.504096111103193,
+    "gpu_batches_per_second": 1.0299283064968987,
+    "producer_headroom": 12.140744197655318,
     "sustainable": true,
-    "buffer_seconds": 60.09220457820395,
+    "buffer_seconds": 60.198364885106386,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368936.0222461823,
+    "generation_bytes_per_second": 382064.1062599746,
     "upload_bytes_per_second": 3566919.5818388737,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.16508198298551,
+    "estimated_full_shards_per_hour": 81.98206320619038,
     "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

@@ -104,3 +104,4 @@
 {"step": 247500, "loss": 0.5994236934930086, "lr": 0.0003, "grad_norm": 0.13597782829188143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36227420155, "tokens_per_second": 150848.63109803176, "data_wait_s": 1.4650999219156802e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 247600, "loss": 0.6316686728969216, "lr": 0.0003, "grad_norm": 0.17470536394397362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36242260478, "tokens_per_second": 151218.82265771617, "data_wait_s": 1.1083999197580852e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 247700, "loss": 0.6217366997152567, "lr": 0.0003, "grad_norm": 0.14484045002700272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36257059647, "tokens_per_second": 150757.91440468692, "data_wait_s": 1.2986998626729473e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 247800, "loss": 0.5763643234968185, "lr": 0.0003, "grad_norm": 0.15381343837353498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36271597498, "tokens_per_second": 148702.6365903454, "data_wait_s": 1.8566999642644078e-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:c33f0cc85c2d3203b343ff1f85ef6f47609e87cfa43cddf8afa3880c9a9750ae
-size 226239
+oid sha256:c01d40070242611b709bbf35f48fd90d16d5261e28ba05f6df2c6a301750b9f5
+size 228357