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": 75429.36580886084,
-      "required_pages_per_second_upper_estimate": 0.006538026254332212
+      "required_token_positions_per_second": 75069.84228901731,
+      "required_pages_per_second_upper_estimate": 0.006506863666836206
     },
     {
       "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": 9428.670726107604,
-      "required_pages_per_second_upper_estimate": 0.003761237723834213
+      "required_token_positions_per_second": 9383.730286127164,
+      "required_pages_per_second_upper_estimate": 0.0037433103104065595
     },
     {
       "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": 18857.34145221521,
-      "required_pages_per_second_upper_estimate": 0.004362314067095466
+      "required_token_positions_per_second": 18767.460572254327,
+      "required_pages_per_second_upper_estimate": 0.004341521707365977
     },
     {
       "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": 28286.012178322817,
-      "required_pages_per_second_upper_estimate": 0.0017224697364233209
+      "required_token_positions_per_second": 28151.190858381495,
+      "required_pages_per_second_upper_estimate": 0.001714259825391693
     },
     {
       "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": 28286.012178322817,
-      "required_pages_per_second_upper_estimate": 0.0021775355181804673
+      "required_token_positions_per_second": 28151.190858381495,
+      "required_pages_per_second_upper_estimate": 0.0021671565997620846
     },
     {
       "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": 28286.012178322817,
-      "required_pages_per_second_upper_estimate": 0.35121760406177055
+      "required_token_positions_per_second": 28151.190858381495,
+      "required_pages_per_second_upper_estimate": 0.3495435744860312
     }
   ],
   "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": 0.20222008119542068,
-    "producer_p95_seconds": 0.2778586259992153,
-    "gpu_update_seconds": 0.9774442514448268,
-    "producer_batches_per_second": 4.945107301354626,
-    "gpu_batches_per_second": 1.0230762506627176,
-    "producer_headroom": 4.83356670448692,
+    "producer_batch_seconds": 0.08968605823433506,
+    "producer_p95_seconds": 0.28383041599954595,
+    "gpu_update_seconds": 0.9821254148390075,
+    "producer_batches_per_second": 11.150005025164143,
+    "gpu_batches_per_second": 1.018199900838451,
+    "producer_headroom": 10.950703310796353,
     "sustainable": true,
-    "buffer_seconds": 60.601543589579265,
+    "buffer_seconds": 60.891775720018465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 461558.14343787107,
-    "upload_bytes_per_second": 5153822.140038256,
+    "generation_bytes_per_second": 465038.2789360121,
+    "upload_bytes_per_second": 5129245.098199123,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.03963305809116,
+    "estimated_full_shards_per_hour": 99.78638912258408,
     "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": 0.8949287559998993,
-      "bytes": 6264749,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0034355950001554,
       "bytes": 860910,
@@ -562,6 +556,12 @@
       "bytes": 11249015,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.3934125020005013,
+      "bytes": 6804482,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -138,3 +138,4 @@
 {"step": 168800, "loss": 0.6158970892429352, "lr": 0.0003, "grad_norm": 0.18008997262078796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24704702692, "tokens_per_second": 149343.14134793746, "data_wait_s": 1.2721000530291349e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 168900, "loss": 0.6203293735161424, "lr": 0.0003, "grad_norm": 0.14527842049973017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24719258931, "tokens_per_second": 148840.7846523434, "data_wait_s": 1.0940999345621094e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 169000, "loss": 0.6521192621439695, "lr": 0.0003, "grad_norm": 0.16166802150130857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24733838653, "tokens_per_second": 149149.88780075763, "data_wait_s": 1.673299993854016e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 169100, "loss": 0.6279681622982025, "lr": 0.0003, "grad_norm": 0.12974266857108036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24748531140, "tokens_per_second": 150551.60547993824, "data_wait_s": 1.5487999917240813e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:416c6090465f382f8373cf2b5b8a952a089de97418ec69bf9693e7af1fa6df5a
-size 298251
+oid sha256:a4eb9d447c03e9676782ff2785dfbf8da9a6b3044953298df6158fe56b13992c
+size 300369