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": 75862.88643226739,
-      "required_pages_per_second_upper_estimate": 0.006575602723221117
+      "required_token_positions_per_second": 75602.35161990508,
+      "required_pages_per_second_upper_estimate": 0.006553020226004997
     },
     {
       "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": 9482.860804033424,
-      "required_pages_per_second_upper_estimate": 0.0037828549561326886
+      "required_token_positions_per_second": 9450.293952488135,
+      "required_pages_per_second_upper_estimate": 0.003769863552133451
     },
     {
       "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": 18965.721608066848,
-      "required_pages_per_second_upper_estimate": 0.004387385908726147
+      "required_token_positions_per_second": 18900.58790497627,
+      "required_pages_per_second_upper_estimate": 0.004372318372830164
     },
     {
       "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": 28448.58241210027,
-      "required_pages_per_second_upper_estimate": 0.0017323694107202697
+      "required_token_positions_per_second": 28350.881857464406,
+      "required_pages_per_second_upper_estimate": 0.0017264199595381404
     },
     {
       "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": 28448.58241210027,
-      "required_pages_per_second_upper_estimate": 0.0021900506247998657
+      "required_token_positions_per_second": 28350.881857464406,
+      "required_pages_per_second_upper_estimate": 0.002182529365651552
     },
     {
       "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": 28448.58241210027,
-      "required_pages_per_second_upper_estimate": 0.3532361822777142
+      "required_token_positions_per_second": 28350.881857464406,
+      "required_pages_per_second_upper_estimate": 0.352023068371859
     }
   ],
   "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": 0.10772339923448726,
-    "producer_p95_seconds": 0.34232484200038016,
-    "gpu_update_seconds": 0.9718586184540516,
-    "producer_batches_per_second": 9.283034207110813,
-    "gpu_batches_per_second": 1.0289562504376544,
-    "producer_headroom": 9.021796799584417,
+    "producer_batch_seconds": 0.14335885924975855,
+    "producer_p95_seconds": 0.3809784810000565,
+    "gpu_update_seconds": 0.9752077603441691,
+    "producer_batches_per_second": 6.9755019343297695,
+    "gpu_batches_per_second": 1.0254225208862995,
+    "producer_headroom": 6.802563618654155,
     "sustainable": true,
-    "buffer_seconds": 60.255234344151205,
+    "buffer_seconds": 60.462881141338485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278782.0156129428,
-    "upload_bytes_per_second": 5205347.348566171,
+    "generation_bytes_per_second": 284435.6098417257,
+    "upload_bytes_per_second": 5270113.022604552,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.82013083735669,
+    "estimated_full_shards_per_hour": 61.033260549915575,
     "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": 1.2683707119977043,
-      "bytes": 2325676,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.4917878320011368,
       "bytes": 10275338,
@@ -562,6 +556,12 @@
       "bytes": 10043413,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0213671140008955,
+      "bytes": 6189327,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -355,3 +355,4 @@
 {"step": 390500, "loss": 0.559561824426055, "lr": 0.0003, "grad_norm": 0.1650607503399263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57158681334, "tokens_per_second": 148375.22940628073, "data_wait_s": 9.72800626186654e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 390600, "loss": 0.5853303074836731, "lr": 0.0003, "grad_norm": 0.1554799104372895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57173100140, "tokens_per_second": 147704.57057288743, "data_wait_s": 1.0850002581719309e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 390700, "loss": 0.6517350496724248, "lr": 0.0003, "grad_norm": 0.16227614378862704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57187738408, "tokens_per_second": 149194.12123201502, "data_wait_s": 1.2413998774718493e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 390800, "loss": 0.5534218586981297, "lr": 0.0003, "grad_norm": 0.150613119189268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57202463395, "tokens_per_second": 150439.76599649354, "data_wait_s": 1.170600444311276e-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 sha256:78eefd8bd5aa8a950b6162b9575c4313cfa4b76b95f0c4c2b94d97848945439e
-size 761772
+oid sha256:70e265d8a31c5f38e9fbdc3367ad664049f126a3f334fcde7c373b2f916a1096
+size 763890