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": 74704.34740469298,
-      "required_pages_per_second_upper_estimate": 0.006475183496601288
+      "required_token_positions_per_second": 74778.77638337637,
+      "required_pages_per_second_upper_estimate": 0.006481634811835043
     },
     {
       "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": 9338.043425586622,
-      "required_pages_per_second_upper_estimate": 0.0037250851386575004
+      "required_token_positions_per_second": 9347.347047922047,
+      "required_pages_per_second_upper_estimate": 0.003728796492708651
     },
     {
       "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": 18676.086851173244,
-      "required_pages_per_second_upper_estimate": 0.004320384005116435
+      "required_token_positions_per_second": 18694.694095844094,
+      "required_pages_per_second_upper_estimate": 0.004324688463694715
     },
     {
       "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": 28014.130276759865,
-      "required_pages_per_second_upper_estimate": 0.0017059135550722324
+      "required_token_positions_per_second": 28042.04114376614,
+      "required_pages_per_second_upper_estimate": 0.0017076131804359125
     },
     {
       "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": 28014.130276759865,
-      "required_pages_per_second_upper_estimate": 0.0021566052967808783
+      "required_token_positions_per_second": 28042.04114376614,
+      "required_pages_per_second_upper_estimate": 0.002158753952585246
     },
     {
       "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": 28014.130276759865,
-      "required_pages_per_second_upper_estimate": 0.34784174077454916
+      "required_token_positions_per_second": 28042.04114376614,
+      "required_pages_per_second_upper_estimate": 0.3481883003311042
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.18953999292186552,
+    "producer_batch_seconds": 0.23463934814839504,
     "producer_p95_seconds": 0.17552479299956758,
-    "gpu_update_seconds": 0.9869305142389659,
-    "producer_batches_per_second": 5.275931398880194,
-    "gpu_batches_per_second": 1.0132425591999372,
-    "producer_headroom": 5.206977688586337,
+    "gpu_update_seconds": 0.985948200355817,
+    "producer_batches_per_second": 4.261859777105932,
+    "gpu_batches_per_second": 1.0142520668318193,
+    "producer_headroom": 4.201972977406436,
     "sustainable": true,
-    "buffer_seconds": 60.20276136857692,
+    "buffer_seconds": 60.14284022170484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 155,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 636967.1548203713,
+    "generation_bytes_per_second": 564117.4594634526,
     "upload_bytes_per_second": 4560660.040824975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.33916179398706,
+    "estimated_full_shards_per_hour": 60.52323740924684,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -22,3 +22,4 @@
 {"step": 239300, "loss": 0.5887480089440942, "lr": 0.0003, "grad_norm": 0.1651936403064816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35026179444, "tokens_per_second": 149111.379039463, "data_wait_s": 1.3530999694921775e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239400, "loss": 0.6398460641503334, "lr": 0.0003, "grad_norm": 0.17507099564559192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35040837015, "tokens_per_second": 148336.3790418062, "data_wait_s": 1.5294999684556387e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239500, "loss": 0.6293351035565138, "lr": 0.0003, "grad_norm": 0.14969589073221598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35055559803, "tokens_per_second": 150727.93092820636, "data_wait_s": 1.1714999800460646e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 239600, "loss": 0.54215855197981, "lr": 0.0003, "grad_norm": 0.16648286641590115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35070155866, "tokens_per_second": 147424.67963321402, "data_wait_s": 1.2212000001454726e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3982ca76c967ab40c6d94c3a525793f453509f82766cf5c7a6701aaf20889a43
-size 50997
+oid sha256:cbf1b28109e247cc5e801ea5a9efdaeb4daf8271b350f2aa5884aecec1f2ea11
+size 53115