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": 76366.52613653135,
-      "required_pages_per_second_upper_estimate": 0.005283213472878548
+      "required_token_positions_per_second": 76650.02015731245,
+      "required_pages_per_second_upper_estimate": 0.00530282624703294
     },
     {
       "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": 9545.815767066419,
-      "required_pages_per_second_upper_estimate": 0.003996633722369866
+      "required_token_positions_per_second": 9581.252519664056,
+      "required_pages_per_second_upper_estimate": 0.004011470350678953
     },
     {
       "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": 19091.631534132837,
-      "required_pages_per_second_upper_estimate": 0.0039762833683787396
+      "required_token_positions_per_second": 19162.505039328113,
+      "required_pages_per_second_upper_estimate": 0.0039910444504508945
     },
     {
       "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": 28637.447301199256,
-      "required_pages_per_second_upper_estimate": 0.0017167712388567332
+      "required_token_positions_per_second": 28743.75755899217,
+      "required_pages_per_second_upper_estimate": 0.0017231443764850537
     },
     {
       "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": 28637.447301199256,
-      "required_pages_per_second_upper_estimate": 0.0022264355648497253
+      "required_token_positions_per_second": 28743.75755899217,
+      "required_pages_per_second_upper_estimate": 0.002234700719783719
     },
     {
       "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": 28637.447301199256,
-      "required_pages_per_second_upper_estimate": 0.3673304254845276
+      "required_token_positions_per_second": 28743.75755899217,
+      "required_pages_per_second_upper_estimate": 0.3686940593244336
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08067848345316264,
-    "producer_p95_seconds": 0.16885126899978786,
-    "gpu_update_seconds": 0.9654491795028874,
-    "producer_batches_per_second": 12.394878500418805,
-    "gpu_batches_per_second": 1.0357873011139778,
-    "producer_headroom": 11.966625278267315,
+    "producer_batch_seconds": 0.07085656189848066,
+    "producer_p95_seconds": 0.16695759999947768,
+    "gpu_update_seconds": 0.9618784163224556,
+    "producer_batches_per_second": 14.113018938637529,
+    "gpu_batches_per_second": 1.039632434859381,
+    "producer_headroom": 13.575008306225492,
     "sustainable": true,
-    "buffer_seconds": 60.823298308681906,
+    "buffer_seconds": 60.598340228314704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 512294.11291257874,
-    "upload_bytes_per_second": 5413574.570446147,
+    "generation_bytes_per_second": 521909.1407248985,
+    "upload_bytes_per_second": 5446789.200541572,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.9263910344412,
+    "estimated_full_shards_per_hour": 111.98955217657296,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 6169276,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.2401527040001383,
+      "bytes": 8401681,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 609400, "loss": 0.527541995048523, "lr": 0.0001, "grad_norm": 0.15548710990944486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89198141513, "tokens_per_second": 149871.48394250727, "data_wait_s": 1.3796000530419406e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 609500, "loss": 0.539238452911377, "lr": 0.0001, "grad_norm": 0.14926697168009725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89212815465, "tokens_per_second": 152162.7101586062, "data_wait_s": 1.1928999811061658e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 609600, "loss": 0.5150657892227173, "lr": 0.0001, "grad_norm": 0.1393956148179828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89227678418, "tokens_per_second": 154392.57602208518, "data_wait_s": 1.1154999810969457e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 609700, "loss": 0.6639874577522278, "lr": 0.0001, "grad_norm": 0.17147602627155376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89242403577, "tokens_per_second": 152148.1796035995, "data_wait_s": 1.30259995785309e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd78ec390a5f676c184918b40d2b5bf6159e1ba19adb10ba0aa3c1ef9a317d8e
-size 111084
+oid sha256:ee33dfa7fed192db0ecaf4202da089991ba0e655d96ec7c549a5cccacc651d1d
+size 113202