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": 75604.30661243413,
-      "required_pages_per_second_upper_estimate": 0.005230481357608542
+      "required_token_positions_per_second": 76399.23722511127,
+      "required_pages_per_second_upper_estimate": 0.005285476501886689
     },
     {
       "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": 9450.538326554266,
-      "required_pages_per_second_upper_estimate": 0.003956743047646633
+      "required_token_positions_per_second": 9549.904653138909,
+      "required_pages_per_second_upper_estimate": 0.003998345653582766
     },
     {
       "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": 18901.076653108532,
-      "required_pages_per_second_upper_estimate": 0.003936595811931524
+      "required_token_positions_per_second": 19099.809306277817,
+      "required_pages_per_second_upper_estimate": 0.003977986582654185
     },
     {
       "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": 28351.614979662798,
-      "required_pages_per_second_upper_estimate": 0.0016996360276213067
+      "required_token_positions_per_second": 28649.713959416724,
+      "required_pages_per_second_upper_estimate": 0.001717506606286768
     },
     {
       "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": 28351.614979662798,
-      "required_pages_per_second_upper_estimate": 0.0022042133590937786
+      "required_token_positions_per_second": 28649.713959416724,
+      "required_pages_per_second_upper_estimate": 0.002227389243570806
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28351.614979662798,
-      "required_pages_per_second_upper_estimate": 0.3636640753666936
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28649.713959416724,
+      "required_pages_per_second_upper_estimate": 0.36748776900522984
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2996538372655664,
-    "producer_p95_seconds": 0.29529358700165176,
-    "gpu_update_seconds": 0.9751825432107654,
-    "producer_batches_per_second": 3.3371840291628105,
-    "gpu_batches_per_second": 1.0254490371695169,
-    "producer_headroom": 3.254363608721339,
+    "producer_batch_seconds": 0.45149098679692656,
+    "producer_p95_seconds": 0.31494102099895827,
+    "gpu_update_seconds": 0.9650358128937802,
+    "producer_batches_per_second": 2.21488363941534,
+    "gpu_batches_per_second": 1.036230973647885,
+    "producer_headroom": 2.137442033428317,
     "sustainable": true,
-    "buffer_seconds": 60.461317679067456,
+    "buffer_seconds": 60.79725621230815,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 489622.771880808,
-    "upload_bytes_per_second": 3255229.205933221,
+    "generation_bytes_per_second": 482638.6223492864,
+    "upload_bytes_per_second": 3258863.7294597602,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.06164901083164,
+    "estimated_full_shards_per_hour": 103.56301310404724,
     "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": [
-    {
-      "seconds": 1.4953640580024512,
-      "bytes": 6379384,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.7474238050017448,
       "bytes": 6290249,
@@ -562,6 +556,12 @@
       "bytes": 6184562,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.9332000800022797,
+      "bytes": 8278775,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -255,3 +255,4 @@
 {"step": 825500, "loss": 0.5523240566253662, "lr": 3e-05, "grad_norm": 0.2925730057574528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120830104871, "tokens_per_second": 152308.17124629006, "data_wait_s": 1.6930996935116127e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 825600, "loss": 0.5206471681594849, "lr": 3e-05, "grad_norm": 0.269644409841144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120844770070, "tokens_per_second": 151627.88770341396, "data_wait_s": 7.342998287640512e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 825700, "loss": 0.5310903191566467, "lr": 3e-05, "grad_norm": 0.28555193636918114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120859220936, "tokens_per_second": 149203.39291495833, "data_wait_s": 1.4158002159092575e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 825800, "loss": 0.5232481360435486, "lr": 3e-05, "grad_norm": 0.29386732985607794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120873850133, "tokens_per_second": 151178.61603742206, "data_wait_s": 2.2290998458629474e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a437d57967c8b3e2abaa60135dc7d1ad83c37f470b8eb2a16b1a34c2c49a8c84
-size 548406
+oid sha256:1bdc7f46f8077ba4dda552442afc881280e67da76f9bd303ff1afd2834b5f12a
+size 550524