Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77057.0906808676,
-      "required_pages_per_second_upper_estimate": 0.005330988330386231
+      "required_token_positions_per_second": 76554.73442421605,
+      "required_pages_per_second_upper_estimate": 0.005296234158923969
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9632.13633510845,
-      "required_pages_per_second_upper_estimate": 0.004032774341630625
+      "required_token_positions_per_second": 9569.341803027006,
+      "required_pages_per_second_upper_estimate": 0.004006483582347067
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19264.2726702169,
-      "required_pages_per_second_upper_estimate": 0.004012239964172125
+      "required_token_positions_per_second": 19138.68360605401,
+      "required_pages_per_second_upper_estimate": 0.00398608307411334
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28896.40900532535,
-      "required_pages_per_second_upper_estimate": 0.0017322955976074655
+      "required_token_positions_per_second": 28708.02540908102,
+      "required_pages_per_second_upper_estimate": 0.0017210022886577142
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28896.40900532535,
-      "required_pages_per_second_upper_estimate": 0.0022465686982933715
+      "required_token_positions_per_second": 28708.02540908102,
+      "required_pages_per_second_upper_estimate": 0.002231922702297244
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28896.40900532535,
-      "required_pages_per_second_upper_estimate": 0.3706521081736426
+      "required_token_positions_per_second": 28708.02540908102,
+      "required_pages_per_second_upper_estimate": 0.3682357256715668
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04123836332031772,
-    "producer_p95_seconds": 0.12829015700003765,
-    "gpu_update_seconds": 0.9567970883477155,
-    "producer_batches_per_second": 24.24926499222413,
-    "gpu_batches_per_second": 1.0451536821949272,
-    "producer_headroom": 23.201626139132234,
+    "producer_batch_seconds": 0.1545397423984305,
+    "producer_p95_seconds": 0.1336200910000116,
+    "gpu_update_seconds": 0.9630756419511282,
+    "producer_batches_per_second": 6.470827403230847,
+    "gpu_batches_per_second": 1.0383400393909512,
+    "producer_headroom": 6.2318962553215,
     "sustainable": true,
-    "buffer_seconds": 60.27821656590608,
+    "buffer_seconds": 60.67376544292107,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 66,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385617.78286591504,
-    "upload_bytes_per_second": 2715257.5497594303,
+    "generation_bytes_per_second": 435030.6625134022,
+    "upload_bytes_per_second": 2480901.3962539616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.74459948046768,
+    "estimated_full_shards_per_hour": 93.34745317985104,
     "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": [
@@ -226,6 +226,12 @@
       "bytes": 6120157,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.5483520990001125,
+      "bytes": 929612,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 843200, "loss": 0.5120399594306946, "lr": 3e-05, "grad_norm": 0.29002387894380416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123420670414, "tokens_per_second": 150754.8370364941, "data_wait_s": 1.2794999975085375e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843300, "loss": 0.5344165563583374, "lr": 3e-05, "grad_norm": 0.298620691567682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123435375990, "tokens_per_second": 151886.9994461236, "data_wait_s": 6.656999858023482e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 843400, "loss": 0.4981638491153717, "lr": 3e-05, "grad_norm": 0.30460754308575466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123450026610, "tokens_per_second": 151846.02516218185, "data_wait_s": 1.2015000038445578e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 843500, "loss": 0.5259416103363037, "lr": 3e-05, "grad_norm": 0.2947109431887774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123464715616, "tokens_per_second": 151901.25280813174, "data_wait_s": 1.4360000022861641e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8cd2ba8f21f2ce14e701eeb8347b1e83568991d2a3d114134ce835d6466870e
-size 19227
+oid sha256:9a407765c8e873e9d8d726b3ae76797f746e7b0ccb3dcd7e9ebdb33175fa094a
+size 21345