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": 75336.58356468836,
-      "required_pages_per_second_upper_estimate": 0.0065299841245617905
+      "required_token_positions_per_second": 75553.26978012547,
+      "required_pages_per_second_upper_estimate": 0.006548765936529686
     },
     {
       "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": 9417.072945586046,
-      "required_pages_per_second_upper_estimate": 0.0037566111957818915
+      "required_token_positions_per_second": 9444.158722515684,
+      "required_pages_per_second_upper_estimate": 0.0037674161171675777
     },
     {
       "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": 18834.14589117209,
-      "required_pages_per_second_upper_estimate": 0.004356948182276073
+      "required_token_positions_per_second": 18888.317445031367,
+      "required_pages_per_second_upper_estimate": 0.004369479817874631
     },
     {
       "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": 28251.218836758137,
-      "required_pages_per_second_upper_estimate": 0.0017203510044685871
+      "required_token_positions_per_second": 28332.47616754705,
+      "required_pages_per_second_upper_estimate": 0.0017252991495893654
     },
     {
       "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": 28251.218836758137,
-      "required_pages_per_second_upper_estimate": 0.0021748570304326847
+      "required_token_positions_per_second": 28332.47616754705,
+      "required_pages_per_second_upper_estimate": 0.002181112444691503
     },
     {
       "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": 28251.218836758137,
-      "required_pages_per_second_upper_estimate": 0.35078558720536074
+      "required_token_positions_per_second": 28332.47616754705,
+      "required_pages_per_second_upper_estimate": 0.35179453130296695
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.09920663029711818,
-    "producer_p95_seconds": 0.3029395539997495,
-    "gpu_update_seconds": 0.9786480420457726,
-    "producer_batches_per_second": 10.079971439459815,
-    "gpu_batches_per_second": 1.021817810935986,
-    "producer_headroom": 9.864744313104655,
+    "producer_batch_seconds": 0.09391089656259055,
+    "producer_p95_seconds": 0.2890647509993869,
+    "gpu_update_seconds": 0.9758412867446061,
+    "producer_batches_per_second": 10.648391577579194,
+    "gpu_batches_per_second": 1.0247568058285246,
+    "producer_headroom": 10.391140138825307,
     "sustainable": true,
-    "buffer_seconds": 60.6761786068379,
+    "buffer_seconds": 60.50215977816558,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277106.75297833997,
-    "upload_bytes_per_second": 5003427.660391841,
+    "generation_bytes_per_second": 275127.41273950133,
+    "upload_bytes_per_second": 4935518.699345201,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.460658474089136,
+    "estimated_full_shards_per_hour": 59.03593813551693,
     "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": 1.1773780209987308,
-      "bytes": 12342839,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.146244535997539,
       "bytes": 6228835,
@@ -562,6 +556,12 @@
       "bytes": 1234734,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.0084878679990652,
+      "bytes": 5849421,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -251,3 +251,4 @@
 {"step": 380100, "loss": 0.6261854413896799, "lr": 0.0003, "grad_norm": 0.18086174655739745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55635464482, "tokens_per_second": 134259.735991459, "data_wait_s": 1.5268000424839556e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 380200, "loss": 0.5697666611522436, "lr": 0.0003, "grad_norm": 0.17639096696056597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55649946284, "tokens_per_second": 147124.3801264208, "data_wait_s": 1.1393000022508204e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 380300, "loss": 0.6201922632753849, "lr": 0.0003, "grad_norm": 0.15635784513159612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55664393099, "tokens_per_second": 147345.95866581547, "data_wait_s": 1.7039001249941066e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 380400, "loss": 0.6774350851774216, "lr": 0.0003, "grad_norm": 0.2429773700394193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55678942383, "tokens_per_second": 147843.0419820773, "data_wait_s": 1.2770000466844067e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9a1d550b8432a7ee20b7ab661aea910c45ef67aa0227ac75085a6eb502791f9
-size 539934
+oid sha256:b4f18a64dfe0c236b862d3dc041880e85bcc8ed98c5ff54c5c253fbcb04b7960
+size 542052