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": 77131.22972712477,
-      "required_pages_per_second_upper_estimate": 0.0053361174416325895
+      "required_token_positions_per_second": 77162.75282033582,
+      "required_pages_per_second_upper_estimate": 0.005338298282364584
     },
     {
       "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": 9641.403715890596,
-      "required_pages_per_second_upper_estimate": 0.004036654400439193
+      "required_token_positions_per_second": 9645.344102541978,
+      "required_pages_per_second_upper_estimate": 0.004038304158045496
     },
     {
       "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": 19282.80743178119,
-      "required_pages_per_second_upper_estimate": 0.004016100266211435
+      "required_token_positions_per_second": 19290.688205083956,
+      "required_pages_per_second_upper_estimate": 0.004017741623460432
     },
     {
       "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": 28924.211147671787,
-      "required_pages_per_second_upper_estimate": 0.001733962293589721
+      "required_token_positions_per_second": 28936.03230762593,
+      "required_pages_per_second_upper_estimate": 0.0017346709540791057
     },
     {
       "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": 28924.211147671787,
-      "required_pages_per_second_upper_estimate": 0.0022487301925721067
+      "required_token_positions_per_second": 28936.03230762593,
+      "required_pages_per_second_upper_estimate": 0.0022496492357627533
     },
     {
       "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": 28924.211147671787,
-      "required_pages_per_second_upper_estimate": 0.371008724204048
+      "required_token_positions_per_second": 28936.03230762593,
+      "required_pages_per_second_upper_estimate": 0.3711603533513671
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12019644846884603,
-    "producer_p95_seconds": 0.3451233039959334,
-    "gpu_update_seconds": 0.9558774086817399,
-    "producer_batches_per_second": 8.319713375384731,
-    "gpu_batches_per_second": 1.046159257366601,
-    "producer_headroom": 7.952626062237569,
+    "producer_batch_seconds": 0.11016767853908505,
+    "producer_p95_seconds": 0.3383885450020898,
+    "gpu_update_seconds": 0.9554869066383204,
+    "producer_batches_per_second": 9.077072452291189,
+    "gpu_batches_per_second": 1.0465868166820722,
+    "producer_headroom": 8.67302387877162,
     "sustainable": true,
-    "buffer_seconds": 60.220276746949615,
+    "buffer_seconds": 60.19567511821418,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 433949.75347783306,
-    "upload_bytes_per_second": 5802477.509922484,
+    "generation_bytes_per_second": 434154.724387061,
+    "upload_bytes_per_second": 5890122.429338022,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.11551526309246,
+    "estimated_full_shards_per_hour": 93.15949724873421,
     "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": 0.9428314830001909,
-      "bytes": 5761904,
-      "pages": 20,
-      "shards": 2
-    },
     {
       "seconds": 1.1127153689994884,
       "bytes": 6912505,
@@ -562,6 +556,12 @@
       "bytes": 5087118,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.1673960980042466,
+      "bytes": 13547167,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 517900, "loss": 0.6210393905639648, "lr": 0.0003, "grad_norm": 0.17097218424655108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75802622273, "tokens_per_second": 147137.5662933031, "data_wait_s": 1.4174998796079308e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 518000, "loss": 0.5508511066436768, "lr": 0.0003, "grad_norm": 0.1640645318245509, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75817400631, "tokens_per_second": 153574.79664636852, "data_wait_s": 1.699499989626929e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 518100, "loss": 0.6016225814819336, "lr": 0.0003, "grad_norm": 0.19384700074382744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75832116061, "tokens_per_second": 153208.25223069056, "data_wait_s": 9.808005415834486e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 518200, "loss": 0.5854417085647583, "lr": 0.0003, "grad_norm": 0.18373835397396274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75846794304, "tokens_per_second": 152599.15833331883, "data_wait_s": 1.7915997887030244e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9fb33605547c31c462d56dcf4000d789628e4c470dc860a7b5a6e84b8038cb7d
-size 705921
+oid sha256:3df4c222bb03d584df80a6d977e07e9195d32050374d3272cb1726044b9f7075
+size 708039