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": 75476.42851012785,
-      "required_pages_per_second_upper_estimate": 0.006542105529998711
+      "required_token_positions_per_second": 75756.8246552545,
+      "required_pages_per_second_upper_estimate": 0.0065664095572007565
     },
     {
       "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": 9434.553563765981,
-      "required_pages_per_second_upper_estimate": 0.0037635844757324005
+      "required_token_positions_per_second": 9469.603081906813,
+      "required_pages_per_second_upper_estimate": 0.0037775662525557736
     },
     {
       "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": 18869.107127531963,
-      "required_pages_per_second_upper_estimate": 0.004365035849011184
+      "required_token_positions_per_second": 18939.206163813626,
+      "required_pages_per_second_upper_estimate": 0.004381252027354045
     },
     {
       "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": 28303.66069129794,
-      "required_pages_per_second_upper_estimate": 0.0017235444382689145
+      "required_token_positions_per_second": 28408.809245720437,
+      "required_pages_per_second_upper_estimate": 0.0017299474335614148
     },
     {
       "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": 28303.66069129794,
-      "required_pages_per_second_upper_estimate": 0.0021788941495635033
+      "required_token_positions_per_second": 28408.809245720437,
+      "required_pages_per_second_upper_estimate": 0.002186988776352775
     },
     {
       "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": 28303.66069129794,
-      "required_pages_per_second_upper_estimate": 0.35143673952714827
+      "required_token_positions_per_second": 28408.809245720437,
+      "required_pages_per_second_upper_estimate": 0.3527423326635017
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.47375192532805954,
-    "producer_p95_seconds": 0.33176186800119467,
-    "gpu_update_seconds": 0.9768347741852513,
-    "producer_batches_per_second": 2.1108093635873435,
-    "gpu_batches_per_second": 1.0237145794016906,
-    "producer_headroom": 2.0619119880279566,
+    "producer_batch_seconds": 0.10067332883593849,
+    "producer_p95_seconds": 0.31596500299929176,
+    "gpu_update_seconds": 0.9732192490315289,
+    "producer_batches_per_second": 9.93311745586204,
+    "gpu_batches_per_second": 1.0275176955194025,
+    "producer_headroom": 9.667101110936027,
     "sustainable": true,
-    "buffer_seconds": 60.56375599948558,
+    "buffer_seconds": 60.33959343995479,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321002.69101570826,
-    "upload_bytes_per_second": 3616609.995560743,
+    "generation_bytes_per_second": 333926.51210755773,
+    "upload_bytes_per_second": 3671865.947815053,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.87970493176876,
+    "estimated_full_shards_per_hour": 71.65285608692216,
     "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": 1.3160915080006816,
-      "bytes": 2528581,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 8.146259255998302,
       "bytes": 6143251,
@@ -562,6 +556,12 @@
       "bytes": 9738932,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.1697420529999363,
+      "bytes": 8456264,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -237,3 +237,4 @@
 {"step": 303900, "loss": 0.6230304967612028, "lr": 0.0003, "grad_norm": 0.16112688015006368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44483237741, "tokens_per_second": 150450.92052689544, "data_wait_s": 1.2651002180064097e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 304000, "loss": 0.5920179393142462, "lr": 0.0003, "grad_norm": 0.15579633301433685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44497869499, "tokens_per_second": 149782.44256119436, "data_wait_s": 8.88900103745982e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 304100, "loss": 0.5965831391513348, "lr": 0.0003, "grad_norm": 0.16153386027744343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44512430630, "tokens_per_second": 148840.87558526438, "data_wait_s": 1.4952998753869906e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 304200, "loss": 0.5576318129897118, "lr": 0.0003, "grad_norm": 0.1558965124334837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44526920027, "tokens_per_second": 148738.75476426695, "data_wait_s": 1.0335999832022935e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b483d2871174562a3a9af805e791f954345a2ef09c1651630488c712c66bdd07
-size 509499
+oid sha256:e503c4ca0872e38e239dd3b1e100b03572a0faf032b4318922f5c2bc1ca25cd7
+size 511617