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": 77018.60201956956,
-      "required_pages_per_second_upper_estimate": 0.005328325595491634
+      "required_token_positions_per_second": 77233.70057445734,
+      "required_pages_per_second_upper_estimate": 0.0053432066126161805
     },
     {
       "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": 9627.325252446195,
-      "required_pages_per_second_upper_estimate": 0.00403076004178677
+      "required_token_positions_per_second": 9654.212571807167,
+      "required_pages_per_second_upper_estimate": 0.004042017200932134
     },
     {
       "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": 19254.65050489239,
-      "required_pages_per_second_upper_estimate": 0.004010235920888593
+      "required_token_positions_per_second": 19308.425143614335,
+      "required_pages_per_second_upper_estimate": 0.004021435760001786
     },
     {
       "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": 28881.975757338587,
-      "required_pages_per_second_upper_estimate": 0.0017314303464289525
+      "required_token_positions_per_second": 28962.637715421504,
+      "required_pages_per_second_upper_estimate": 0.0017362659076523449
     },
     {
       "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": 28881.975757338587,
-      "required_pages_per_second_upper_estimate": 0.002245446576747549
+      "required_token_positions_per_second": 28962.637715421504,
+      "required_pages_per_second_upper_estimate": 0.002251717689194038
     },
     {
       "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": 28881.975757338587,
-      "required_pages_per_second_upper_estimate": 0.3704669739656827
+      "required_token_positions_per_second": 28962.637715421504,
+      "required_pages_per_second_upper_estimate": 0.3715016189559075
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.0920828439766126,
-    "producer_p95_seconds": 0.25848438600223744,
-    "gpu_update_seconds": 0.9572752304860914,
-    "producer_batches_per_second": 10.859786218744311,
-    "gpu_batches_per_second": 1.0446316463157763,
-    "producer_headroom": 10.39580435557814,
+    "producer_batch_seconds": 0.0770007810702964,
+    "producer_p95_seconds": 0.26235660599922994,
+    "gpu_update_seconds": 0.9546091855189864,
+    "producer_batches_per_second": 12.986881251075479,
+    "gpu_batches_per_second": 1.0475491071839376,
+    "producer_headroom": 12.39739613352096,
     "sustainable": true,
-    "buffer_seconds": 60.30833952062376,
+    "buffer_seconds": 60.14037868769614,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435496.612029153,
-    "upload_bytes_per_second": 4912447.080293131,
+    "generation_bytes_per_second": 436054.330760626,
+    "upload_bytes_per_second": 4858897.732117871,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.44743509918158,
+    "estimated_full_shards_per_hour": 93.56710855592809,
     "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": 1.4081739580014982,
-      "bytes": 5891309,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.286470338000072,
       "bytes": 2778002,
@@ -562,6 +556,12 @@
       "bytes": 7296963,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.3247330819976924,
+      "bytes": 862254,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 865200, "loss": 0.512775719165802, "lr": 3e-05, "grad_norm": 0.31851739616944663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126641287794, "tokens_per_second": 149899.46214735997, "data_wait_s": 1.1513999197632074e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 865300, "loss": 0.5320670008659363, "lr": 3e-05, "grad_norm": 0.3112138196399291, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126655875170, "tokens_per_second": 151309.34469403807, "data_wait_s": 1.1741998605430126e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 865400, "loss": 0.5289158821105957, "lr": 3e-05, "grad_norm": 0.29619837174964825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126670477838, "tokens_per_second": 151739.60537559004, "data_wait_s": 1.4899000234436244e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 865500, "loss": 0.5028296709060669, "lr": 3e-05, "grad_norm": 0.31073990264405715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126685154588, "tokens_per_second": 152093.12917937845, "data_wait_s": 1.0292002116329968e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:953f8cf09a5047914bdb0367d1a737df9cae0607696a4fce7b3afa6f7a054e7c
-size 489102
+oid sha256:0de46aeb0673f77573492711c81890103e594b56fb8d5dce35525d1ce975dbaa
+size 491220