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": 76106.74271288363,
-      "required_pages_per_second_upper_estimate": 0.0065967395675768405
+      "required_token_positions_per_second": 76109.52670539392,
+      "required_pages_per_second_upper_estimate": 0.006596980876991667
     },
     {
       "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": 9513.342839110453,
-      "required_pages_per_second_upper_estimate": 0.003795014695671954
+      "required_token_positions_per_second": 9513.69083817424,
+      "required_pages_per_second_upper_estimate": 0.0037951535177015478
     },
     {
       "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": 19026.685678220907,
-      "required_pages_per_second_upper_estimate": 0.0044014888734252985
+      "required_token_positions_per_second": 19027.38167634848,
+      "required_pages_per_second_upper_estimate": 0.004401649880342964
     },
     {
       "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": 28540.028517331357,
-      "required_pages_per_second_upper_estimate": 0.0017379379987482082
+      "required_token_positions_per_second": 28541.07251452272,
+      "required_pages_per_second_upper_estimate": 0.0017380015726997323
     },
     {
       "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": 28540.028517331357,
-      "required_pages_per_second_upper_estimate": 0.0021970903991195756
+      "required_token_positions_per_second": 28541.07251452272,
+      "required_pages_per_second_upper_estimate": 0.002197170768913336
     },
     {
       "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": 28540.028517331357,
-      "required_pages_per_second_upper_estimate": 0.35437163685425777
+      "required_token_positions_per_second": 28541.07251452272,
+      "required_pages_per_second_upper_estimate": 0.35438459980534065
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05669148961713688,
-    "producer_p95_seconds": 0.18506762800006982,
-    "gpu_update_seconds": 0.968744652206999,
-    "producer_batches_per_second": 17.639331877737728,
-    "gpu_batches_per_second": 1.0322637629243112,
-    "producer_headroom": 17.088008425062867,
+    "producer_batch_seconds": 0.10804525796867637,
+    "producer_p95_seconds": 0.21596477799994318,
+    "gpu_update_seconds": 0.9687092167237832,
+    "producer_batches_per_second": 9.255380743224402,
+    "gpu_batches_per_second": 1.0323015232393924,
+    "producer_headroom": 8.965772630249296,
     "sustainable": true,
-    "buffer_seconds": 60.062168436833936,
+    "buffer_seconds": 60.059971436874555,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328779.06979646796,
+    "generation_bytes_per_second": 330395.48409064265,
     "upload_bytes_per_second": 5126696.548462012,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.5483347932866,
+    "estimated_full_shards_per_hour": 70.89517967261753,
     "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": [

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

@@ -64,3 +64,4 @@
 {"step": 161400, "loss": 0.5813808608800173, "lr": 0.0003, "grad_norm": 0.16874280129162456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23621895067, "tokens_per_second": 148976.63425909603, "data_wait_s": 1.3281000065035187e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 161500, "loss": 0.5784920454025269, "lr": 0.0003, "grad_norm": 0.1443170699606449, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23636375426, "tokens_per_second": 148385.01331216842, "data_wait_s": 1.0714999916672241e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 161600, "loss": 0.5485043972730637, "lr": 0.0003, "grad_norm": 0.14040194826669436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23651055227, "tokens_per_second": 150388.79145948522, "data_wait_s": 1.2374000107229222e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 161700, "loss": 0.5814548656344414, "lr": 0.0003, "grad_norm": 0.1337939706795106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23665732011, "tokens_per_second": 149694.2218500253, "data_wait_s": 1.3785000192001462e-05, "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 sha256:6f14012608c2f4cfa3be47e0317b697599956195fb1c1add54ffd7c48954ee34
-size 140736
+oid sha256:add7bfb7b0935b4ed5a9e869cc1d0dc5e5db4cdefc1622a2be5dab2460f350e9
+size 142854