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": 74773.89887669247,
-      "required_pages_per_second_upper_estimate": 0.00648121204191761
+      "required_token_positions_per_second": 74615.96217969531,
+      "required_pages_per_second_upper_estimate": 0.006467522489308796
     },
     {
       "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": 9346.73735958656,
-      "required_pages_per_second_upper_estimate": 0.0037285532789159723
+      "required_token_positions_per_second": 9326.995272461914,
+      "required_pages_per_second_upper_estimate": 0.003720677865191445
     },
     {
       "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": 18693.47471917312,
-      "required_pages_per_second_upper_estimate": 0.004324406382362181
+      "required_token_positions_per_second": 18653.99054492383,
+      "required_pages_per_second_upper_estimate": 0.004315272413547341
     },
     {
       "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": 28040.212078759676,
-      "required_pages_per_second_upper_estimate": 0.0017075017999733815
+      "required_token_positions_per_second": 27980.985817385743,
+      "required_pages_per_second_upper_estimate": 0.0017038952313918874
     },
     {
       "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": 28040.212078759676,
-      "required_pages_per_second_upper_estimate": 0.0021586131460978727
+      "required_token_positions_per_second": 27980.985817385743,
+      "required_pages_per_second_upper_estimate": 0.0021540537445485226
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28040.212078759676,
-      "required_pages_per_second_upper_estimate": 0.3481655894652107
+      "required_token_positions_per_second": 27980.985817385743,
+      "required_pages_per_second_upper_estimate": 0.34743019751649235
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07027888952347894,
-    "producer_p95_seconds": 0.2698636759996589,
-    "gpu_update_seconds": 0.9860125138262854,
-    "producer_batches_per_second": 14.229023918568286,
-    "gpu_batches_per_second": 1.0141859114134721,
-    "producer_headroom": 14.02999564324186,
+    "producer_batch_seconds": 0.15838886547652464,
+    "producer_p95_seconds": 0.25946448700051405,
+    "gpu_update_seconds": 0.9880995680581474,
+    "producer_batches_per_second": 6.313575117742184,
+    "gpu_batches_per_second": 1.012043757862621,
+    "producer_headroom": 6.2384408467437185,
     "sustainable": true,
-    "buffer_seconds": 60.14676334340341,
+    "buffer_seconds": 60.27407365154699,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393578.17469445564,
+    "generation_bytes_per_second": 400354.3176121502,
     "upload_bytes_per_second": 5357125.110547315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.4527142584348,
+    "estimated_full_shards_per_hour": 85.90671678803807,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -138,3 +138,4 @@
 {"step": 208800, "loss": 0.64070232398808, "lr": 0.0003, "grad_norm": 0.1767233183359933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30560615198, "tokens_per_second": 148293.30779086935, "data_wait_s": 1.477899968449492e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 208900, "loss": 0.6065094582736492, "lr": 0.0003, "grad_norm": 0.172042730851505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30575317023, "tokens_per_second": 148050.1568811849, "data_wait_s": 1.7082998965634033e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209000, "loss": 0.6019245516508818, "lr": 0.0003, "grad_norm": 0.151780107319525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30589992924, "tokens_per_second": 147722.52947019503, "data_wait_s": 1.7903999832924455e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 209100, "loss": 0.6039793342351913, "lr": 0.0003, "grad_norm": 0.1505738887339764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30604528995, "tokens_per_second": 146527.98548877737, "data_wait_s": 2.1677000404451974e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a947b2ab5340349172a743416a89f0cb050ac54d41da646f8fc43f62a1c542cf
-size 298251
+oid sha256:37685a5220dbcc80cf995f454732e9fcb1a12f98255b2ae6c6a1ab42a45e68ed
+size 300369