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": 77326.72024573633,
-      "required_pages_per_second_upper_estimate": 0.005349641929310635
+      "required_token_positions_per_second": 77408.04036536583,
+      "required_pages_per_second_upper_estimate": 0.005355267843875277
     },
     {
       "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": 9665.840030717041,
-      "required_pages_per_second_upper_estimate": 0.004046885375168745
+      "required_token_positions_per_second": 9676.005045670729,
+      "required_pages_per_second_upper_estimate": 0.004051141254660204
     },
     {
       "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": 19331.680061434083,
-      "required_pages_per_second_upper_estimate": 0.004026279146110216
+      "required_token_positions_per_second": 19352.010091341457,
+      "required_pages_per_second_upper_estimate": 0.004030513355200979
     },
     {
       "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": 28997.520092151124,
-      "required_pages_per_second_upper_estimate": 0.0017383570528749803
+      "required_token_positions_per_second": 29028.015137012186,
+      "required_pages_per_second_upper_estimate": 0.0017401851842511636
     },
     {
       "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": 28997.520092151124,
-      "required_pages_per_second_upper_estimate": 0.002254429640553406
+      "required_token_positions_per_second": 29028.015137012186,
+      "required_pages_per_second_upper_estimate": 0.002256800496157827
     },
     {
       "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": 28997.520092151124,
-      "required_pages_per_second_upper_estimate": 0.371949052630817
+      "required_token_positions_per_second": 29028.015137012186,
+      "required_pages_per_second_upper_estimate": 0.3723402103232666
     }
   ],
   "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": 0.26717932064826755,
-    "producer_p95_seconds": 0.2603468769993924,
-    "gpu_update_seconds": 0.9534608446562848,
-    "producer_batches_per_second": 3.742804636128504,
-    "gpu_batches_per_second": 1.0488107672219011,
-    "producer_headroom": 3.5686176697465424,
+    "producer_batch_seconds": 0.071407230726237,
+    "producer_p95_seconds": 0.2508910550022847,
+    "gpu_update_seconds": 0.9524591974167536,
+    "producer_batches_per_second": 14.004184027718809,
+    "gpu_batches_per_second": 1.049913741934758,
+    "producer_headroom": 13.338413879517576,
     "sustainable": true,
-    "buffer_seconds": 60.06803321334594,
+    "buffer_seconds": 60.00492943725548,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412804.5727037102,
-    "upload_bytes_per_second": 3794521.742559691,
+    "generation_bytes_per_second": 422143.22666845564,
+    "upload_bytes_per_second": 3832234.922624881,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.57825170358161,
+    "estimated_full_shards_per_hour": 90.58210945167782,
     "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": 1.089157542999601,
-      "bytes": 7603521,
-      "pages": 46,
-      "shards": 3
-    },
     {
       "seconds": 1.0148978149991308,
       "bytes": 6571585,
@@ -562,6 +556,12 @@
       "bytes": 11162167,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.1609297110007901,
+      "bytes": 12171791,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -259,3 +259,4 @@
 {"step": 550900, "loss": 0.5462967753410339, "lr": 0.0003, "grad_norm": 0.1775689481987864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80634410765, "tokens_per_second": 152436.04036302684, "data_wait_s": 9.16099816095084e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 551000, "loss": 0.5720579028129578, "lr": 0.0003, "grad_norm": 0.1548219317217928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80648910070, "tokens_per_second": 150628.9901113007, "data_wait_s": 1.7459002265240997e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 551100, "loss": 0.5425984859466553, "lr": 0.0003, "grad_norm": 0.1645459257452434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80663517080, "tokens_per_second": 152394.21970044757, "data_wait_s": 1.3423999916994944e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 551200, "loss": 0.5516425371170044, "lr": 0.0003, "grad_norm": 0.14363207303726708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80678179255, "tokens_per_second": 152547.54230483074, "data_wait_s": 1.1325999366817996e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ee937353f905e8ee568a0b34004dec2b18d476f348c1c905013d8d80a8dd7c0
-size 556878
+oid sha256:abfc3a2c2afecf4fca668b471506803eb55d3e1ecaaa0b7dffb1b9159f2f5424
+size 558996