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": 77096.68238645294,
-      "required_pages_per_second_upper_estimate": 0.005333727376444039
+      "required_token_positions_per_second": 76518.75856569826,
+      "required_pages_per_second_upper_estimate": 0.005293745265556216
     },
     {
       "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": 9637.085298306618,
-      "required_pages_per_second_upper_estimate": 0.0040348463691755945
+      "required_token_positions_per_second": 9564.844820712282,
+      "required_pages_per_second_upper_estimate": 0.004004600789759562
     },
     {
       "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": 19274.170596613236,
-      "required_pages_per_second_upper_estimate": 0.00401430144121461
+      "required_token_positions_per_second": 19129.689641424564,
+      "required_pages_per_second_upper_estimate": 0.0039842098684678235
     },
     {
       "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": 28911.255894919854,
-      "required_pages_per_second_upper_estimate": 0.0017331856459687682
+      "required_token_positions_per_second": 28694.534462136846,
+      "required_pages_per_second_upper_estimate": 0.0017201935269878941
     },
     {
       "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": 28911.255894919854,
-      "required_pages_per_second_upper_estimate": 0.002247722978654778
+      "required_token_positions_per_second": 28694.534462136846,
+      "required_pages_per_second_upper_estimate": 0.002230873840512739
     },
     {
       "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": 28911.255894919854,
-      "required_pages_per_second_upper_estimate": 0.37084254813201284
+      "required_token_positions_per_second": 28694.534462136846,
+      "required_pages_per_second_upper_estimate": 0.3680626782896172
     }
   ],
   "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": 0.05337549899217464,
-    "producer_p95_seconds": 0.17178940000030707,
-    "gpu_update_seconds": 0.9563057412825214,
-    "producer_batches_per_second": 18.735187846142846,
-    "gpu_batches_per_second": 1.0456906790697285,
-    "producer_headroom": 17.916567701272918,
+    "producer_batch_seconds": 0.0992391345858934,
+    "producer_p95_seconds": 0.18643715200050792,
+    "gpu_update_seconds": 0.9635284390650152,
+    "producer_batches_per_second": 10.076669896133371,
+    "gpu_batches_per_second": 1.0378520855807598,
+    "producer_headroom": 9.709158015994815,
     "sustainable": true,
-    "buffer_seconds": 60.24726170079885,
+    "buffer_seconds": 60.702291661095956,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395074.6757847272,
-    "upload_bytes_per_second": 4483885.078097433,
+    "generation_bytes_per_second": 394730.2254127227,
+    "upload_bytes_per_second": 4414538.735337501,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.77382855564463,
+    "estimated_full_shards_per_hour": 84.69991752420674,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 10527927,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 1.384273941000174,
+      "bytes": 1166498,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -59,3 +59,4 @@
 {"step": 848400, "loss": 0.5915848016738892, "lr": 3e-05, "grad_norm": 0.2935924442510267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124182260231, "tokens_per_second": 152495.19183081103, "data_wait_s": 7.327000275836326e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 848500, "loss": 0.5295246839523315, "lr": 3e-05, "grad_norm": 0.30518167124983603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124196932998, "tokens_per_second": 153319.56940905136, "data_wait_s": 1.4993000149843283e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 848600, "loss": 0.5495550632476807, "lr": 3e-05, "grad_norm": 0.2998802889979366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124211473077, "tokens_per_second": 151325.2296496637, "data_wait_s": 1.6713000150048174e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 848700, "loss": 0.5220390558242798, "lr": 3e-05, "grad_norm": 0.30243580551810806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124226175902, "tokens_per_second": 152910.14577780204, "data_wait_s": 1.1047999578295276e-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 sha256:85f3b04b52a369ad5c3f532d4b10a36df2dbf0808fa73dce82b7df2e272e8581
-size 130146
+oid sha256:f5537f2172544cfaef66675ab65a7b616284c0e633bc3eb5970464e21af5c9e1
+size 132264