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": 75156.2396899193,
-      "required_pages_per_second_upper_estimate": 0.006514352374574179
+      "required_token_positions_per_second": 73399.49469045641,
+      "required_pages_per_second_upper_estimate": 0.006362082170450232
     },
     {
       "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": 9394.529961239912,
-      "required_pages_per_second_upper_estimate": 0.0037476184622785673
+      "required_token_positions_per_second": 9174.936836307052,
+      "required_pages_per_second_upper_estimate": 0.003660019481533051
     },
     {
       "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": 18789.059922479824,
-      "required_pages_per_second_upper_estimate": 0.004346518336905065
+      "required_token_positions_per_second": 18349.873672614103,
+      "required_pages_per_second_upper_estimate": 0.004244920327412636
     },
     {
       "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": 28183.589883719735,
-      "required_pages_per_second_upper_estimate": 0.0017162327560502429
+      "required_token_positions_per_second": 27524.810508921157,
+      "required_pages_per_second_upper_estimate": 0.0016761165484732684
     },
     {
       "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": 28183.589883719735,
-      "required_pages_per_second_upper_estimate": 0.0021696507664188645
+      "required_token_positions_per_second": 27524.810508921157,
+      "required_pages_per_second_upper_estimate": 0.0021189361065288436
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28183.589883719735,
-      "required_pages_per_second_upper_estimate": 0.34994586194816957
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27524.810508921157,
+      "required_pages_per_second_upper_estimate": 0.34176602690590857
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08091448022655356,
-    "producer_p95_seconds": 0.18719405599949823,
-    "gpu_update_seconds": 0.9809963923712529,
-    "producer_batches_per_second": 12.35872735263313,
-    "gpu_batches_per_second": 1.019371740585928,
-    "producer_headroom": 12.123866947233024,
+    "producer_batch_seconds": 0.3709442375077856,
+    "producer_p95_seconds": 0.21131461600089096,
+    "gpu_update_seconds": 1.0044755799876957,
+    "producer_batches_per_second": 2.695822980614469,
+    "gpu_batches_per_second": 0.9955443615784562,
+    "producer_headroom": 2.707888351996877,
     "sustainable": true,
-    "buffer_seconds": 60.82177632701768,
+    "buffer_seconds": 60.26853479926174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397083.56006381963,
-    "upload_bytes_per_second": 43958.880398765425,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 260225.7665281119,
+    "upload_bytes_per_second": 4203699.583367449,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.20488835750524,
+    "estimated_full_shards_per_hour": 55.83839175112265,
     "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": [
@@ -184,6 +184,18 @@
       "bytes": 43764,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 0.9273372809984721,
+      "bytes": 2566992,
+      "pages": 35,
+      "shards": 2
+    },
+    {
+      "seconds": 1.1000417619998188,
+      "bytes": 10096800,
+      "pages": 18,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 89800, "loss": 0.6096045114099979, "lr": 0.0003, "grad_norm": 0.14591806689246076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13141590162, "tokens_per_second": 148641.57574771694, "data_wait_s": 1.707599949440919e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 89900, "loss": 0.5985537357628345, "lr": 0.0003, "grad_norm": 0.12689724016327258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13156147943, "tokens_per_second": 148430.52635961468, "data_wait_s": 2.1118999939062633e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 90000, "loss": 0.6308812461793423, "lr": 0.0003, "grad_norm": 0.13849090878893888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13170804190, "tokens_per_second": 148330.89333950818, "data_wait_s": 7.241998901008628e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 90100, "loss": 0.6027867775410414, "lr": 0.0003, "grad_norm": 0.14209351892134955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13185465865, "tokens_per_second": 133230.76583391437, "data_wait_s": 1.3549000868806615e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1802f74a00a9d017d1050aa62b8be29a8418589d0559d5d97b7037e474f9b136
-size 106848
+oid sha256:dd23a68a374ed446b1b4bc5096dacad768c501a8fc457b81f7782283f38e6cac
+size 108966