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": 74329.01560218727,
-      "required_pages_per_second_upper_estimate": 0.00644265068722985
+      "required_token_positions_per_second": 74859.59955597336,
+      "required_pages_per_second_upper_estimate": 0.006488640359591291
     },
     {
       "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": 9291.126950273409,
-      "required_pages_per_second_upper_estimate": 0.0037063694551912434
+      "required_token_positions_per_second": 9357.44994449667,
+      "required_pages_per_second_upper_estimate": 0.0037328266892040327
     },
     {
       "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": 18582.253900546817,
-      "required_pages_per_second_upper_estimate": 0.00429867740339254
+      "required_token_positions_per_second": 18714.89988899334,
+      "required_pages_per_second_upper_estimate": 0.004329362718329979
     },
     {
       "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": 27873.380850820224,
-      "required_pages_per_second_upper_estimate": 0.0016973426534876753
+      "required_token_positions_per_second": 28072.349833490007,
+      "required_pages_per_second_upper_estimate": 0.0017094588206226907
     },
     {
       "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": 27873.380850820224,
-      "required_pages_per_second_upper_estimate": 0.002145770016352697
+      "required_token_positions_per_second": 28072.349833490007,
+      "required_pages_per_second_upper_estimate": 0.0021610871994200125
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27873.380850820224,
-      "required_pages_per_second_upper_estimate": 0.34609410396240514
+      "required_token_positions_per_second": 28072.349833490007,
+      "required_pages_per_second_upper_estimate": 0.34856463282081535
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.056166969304598524,
-    "producer_p95_seconds": 0.19404527099868574,
-    "gpu_update_seconds": 0.9919141186343173,
-    "producer_batches_per_second": 17.80405837062189,
-    "gpu_batches_per_second": 1.0081517958195971,
-    "producer_headroom": 17.66009686680935,
+    "producer_batch_seconds": 0.32710891848431345,
+    "producer_p95_seconds": 0.23389736100034497,
+    "gpu_update_seconds": 0.9848837081324855,
+    "producer_batches_per_second": 3.0570857090463437,
+    "gpu_batches_per_second": 1.0153483012691698,
+    "producer_headroom": 3.0108739092043915,
     "sustainable": true,
-    "buffer_seconds": 60.50676123669336,
+    "buffer_seconds": 60.077906196081614,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309083.2801523331,
-    "upload_bytes_per_second": 4023124.433886555,
+    "generation_bytes_per_second": 319877.3417543802,
+    "upload_bytes_per_second": 4045039.515985068,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.32207683017249,
+    "estimated_full_shards_per_hour": 68.63823117707781,
     "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": 1.3559997279999152,
-      "bytes": 10442134,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.8416516239994962,
       "bytes": 2569575,
@@ -562,6 +556,12 @@
       "bytes": 7473663,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0918037890005508,
+      "bytes": 11651730,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -121,3 +121,4 @@
 {"step": 327100, "loss": 0.5740393213927746, "lr": 0.0003, "grad_norm": 0.14900409334726034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47878431430, "tokens_per_second": 148157.81349554347, "data_wait_s": 1.923099989653565e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 327200, "loss": 0.5817774944007397, "lr": 0.0003, "grad_norm": 0.13918941547010838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47893097448, "tokens_per_second": 148836.4416416254, "data_wait_s": 1.1533000360941514e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 327300, "loss": 0.5769318398088217, "lr": 0.0003, "grad_norm": 0.15720138966766922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47907710928, "tokens_per_second": 148577.17979066292, "data_wait_s": 1.4534000001731329e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 327400, "loss": 0.6022498216480017, "lr": 0.0003, "grad_norm": 0.1598802955628392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47922314253, "tokens_per_second": 148855.92447491825, "data_wait_s": 1.4339999324874952e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c65e19288045831a0d6689ec443ee310a09cdfc9254ccb7e1ac13e4fe2115e4f
-size 262245
+oid sha256:fcbbecf113ce0b244f777093941e33971b283460ef8d219eca1aef2a21f8dad1
+size 264363