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": 75240.27912302516,
-      "required_pages_per_second_upper_estimate": 0.0065216366996398
+      "required_token_positions_per_second": 72790.00649781169,
+      "required_pages_per_second_upper_estimate": 0.0063092532786454885
     },
     {
       "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": 9405.034890378145,
-      "required_pages_per_second_upper_estimate": 0.0037518090355744952
+      "required_token_positions_per_second": 9098.75081222646,
+      "required_pages_per_second_upper_estimate": 0.0036296277374447345
     },
     {
       "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": 18810.06978075629,
-      "required_pages_per_second_upper_estimate": 0.004351378597856449
+      "required_token_positions_per_second": 18197.50162445292,
+      "required_pages_per_second_upper_estimate": 0.004209671735727004
     },
     {
       "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": 28215.104671134435,
-      "required_pages_per_second_upper_estimate": 0.0017181518412584871
+      "required_token_positions_per_second": 27296.25243667938,
+      "required_pages_per_second_upper_estimate": 0.0016621985610252741
     },
     {
       "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": 28215.104671134435,
-      "required_pages_per_second_upper_estimate": 0.002172076861992563
+      "required_token_positions_per_second": 27296.25243667938,
+      "required_pages_per_second_upper_estimate": 0.0021013410734384338
     },
     {
       "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": 28215.104671134435,
-      "required_pages_per_second_upper_estimate": 0.3503371701346516
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27296.25243667938,
+      "required_pages_per_second_upper_estimate": 0.3389281005833267
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.11026365909373226,
-    "producer_p95_seconds": 0.3541800799976045,
-    "gpu_update_seconds": 0.9799006710148903,
-    "producer_batches_per_second": 9.069171186763592,
-    "gpu_batches_per_second": 1.020511598348323,
-    "producer_headroom": 8.886886931458553,
+    "producer_batch_seconds": 0.09552100020306398,
+    "producer_p95_seconds": 0.3645251919988368,
+    "gpu_update_seconds": 1.012886295074263,
+    "producer_batches_per_second": 10.468902103978634,
+    "gpu_batches_per_second": 0.9872776488961003,
+    "producer_headroom": 10.603807465594075,
     "sustainable": true,
-    "buffer_seconds": 60.7538416029232,
+    "buffer_seconds": 60.773177704455776,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322285.019038651,
-    "upload_bytes_per_second": 4528254.52502443,
+    "generation_bytes_per_second": 308667.15492542734,
+    "upload_bytes_per_second": 4454698.648017714,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.15486267442367,
+    "estimated_full_shards_per_hour": 66.23278604337803,
     "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": 1.756591566001589,
-      "bytes": 11817435,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.9047177330030536,
       "bytes": 2631324,
@@ -562,6 +556,12 @@
       "bytes": 10678100,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.8894862470006046,
+      "bytes": 957902,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -284,3 +284,4 @@
 {"step": 183400, "loss": 0.6113365180790424, "lr": 0.0003, "grad_norm": 0.16286279749599977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26842072776, "tokens_per_second": 148601.73165427157, "data_wait_s": 1.0948999261017889e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 183500, "loss": 0.630158543586731, "lr": 0.0003, "grad_norm": 0.1525393532708313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26856919474, "tokens_per_second": 150832.76571858124, "data_wait_s": 1.4436001947615296e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 183600, "loss": 0.614504911005497, "lr": 0.0003, "grad_norm": 0.14576641525086484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26871718235, "tokens_per_second": 150524.31812238303, "data_wait_s": 2.158799907192588e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 183700, "loss": 0.6800808813422918, "lr": 0.0003, "grad_norm": 0.16859660292366452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26886354630, "tokens_per_second": 149794.669351485, "data_wait_s": 7.031998393358663e-06, "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:fe8465c2b6e6b94b310fcc27e6dabdf9719172bd92459a098646f03a40018e6b
-size 609828
+oid sha256:11cc95c07f5d361f134cedaef672754192360460ede5c39f4e468ecb3d1b186a
+size 611946