Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76108.73656085691,
-      "required_pages_per_second_upper_estimate": 0.006596912389265803
+      "required_token_positions_per_second": 75971.38435204604,
+      "required_pages_per_second_upper_estimate": 0.006585007047922067
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9513.592070107114,
-      "required_pages_per_second_upper_estimate": 0.003795114117642857
+      "required_token_positions_per_second": 9496.423044005754,
+      "required_pages_per_second_upper_estimate": 0.003788265136431209
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19027.184140214227,
-      "required_pages_per_second_upper_estimate": 0.004401604183834822
+      "required_token_positions_per_second": 18992.84608801151,
+      "required_pages_per_second_upper_estimate": 0.004393660679786801
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28540.776210321343,
-      "required_pages_per_second_upper_estimate": 0.0017379835293284565
+      "required_token_positions_per_second": 28489.269132017263,
+      "required_pages_per_second_upper_estimate": 0.0017348470184964928
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28540.776210321343,
-      "required_pages_per_second_upper_estimate": 0.0021971479585956915
+      "required_token_positions_per_second": 28489.269132017263,
+      "required_pages_per_second_upper_estimate": 0.002193182801127125
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28540.776210321343,
-      "required_pages_per_second_upper_estimate": 0.35438092069882593
+      "required_token_positions_per_second": 28489.269132017263,
+      "required_pages_per_second_upper_estimate": 0.35374137516939125
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09583441909370549,
-    "producer_p95_seconds": 0.3272251869966567,
-    "gpu_update_seconds": 0.9687192736545658,
-    "producer_batches_per_second": 10.434664387355598,
-    "gpu_batches_per_second": 1.03229080621822,
-    "producer_headroom": 10.108260506148278,
+    "producer_batch_seconds": 0.1080072189297141,
+    "producer_p95_seconds": 0.3288641129984171,
+    "gpu_update_seconds": 0.9704706664070994,
+    "producer_batches_per_second": 9.258640393756938,
+    "gpu_batches_per_second": 1.0304278476568745,
+    "producer_headroom": 8.985238912952985,
     "sustainable": true,
-    "buffer_seconds": 60.06059496658308,
+    "buffer_seconds": 60.16918131724017,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 502311.655093655,
-    "upload_bytes_per_second": 4552632.234319538,
+    "generation_bytes_per_second": 482653.6395228164,
+    "upload_bytes_per_second": 4603638.36055719,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.78438796622503,
+    "estimated_full_shards_per_hour": 103.56623543990487,
     "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": 0.8101317319997179,
-      "bytes": 713378,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 1.03739196099923,
       "bytes": 7802740,
@@ -562,6 +556,12 @@
       "bytes": 318780,
       "pages": 9,
       "shards": 1
+    },
+    {
+      "seconds": 0.9711113170014869,
+      "bytes": 6125935,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -250,3 +250,4 @@
 {"step": 65000, "loss": 0.6491995714604855, "lr": 0.0003, "grad_norm": 0.12935625010639407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9512775355, "tokens_per_second": 148919.8180950165, "data_wait_s": 1.1274001735728234e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 65100, "loss": 0.5611817017197609, "lr": 0.0003, "grad_norm": 0.12788663244318282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9527447309, "tokens_per_second": 133901.81024125547, "data_wait_s": 9.081999451154843e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 65200, "loss": 0.5868961215019226, "lr": 0.0003, "grad_norm": 0.13932515037847346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9542147796, "tokens_per_second": 149279.75819578778, "data_wait_s": 1.1594001989578828e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 65300, "loss": 0.6247148029506207, "lr": 0.0003, "grad_norm": 0.12495317407980254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9556871900, "tokens_per_second": 150364.37765044507, "data_wait_s": 9.878000128082931e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:549555e87fff747f472c6cdd231c3da095c2aa228d193438976ed727d087f4e1
-size 537816
+oid sha256:a0795c4d20675edf0fc24b34207f15cdf70e77a0261e4909a4f9fb3afc694d33
+size 539934