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": 74727.34272114724,
-      "required_pages_per_second_upper_estimate": 0.006477176672350447
+      "required_token_positions_per_second": 74486.26450250496,
+      "required_pages_per_second_upper_estimate": 0.006456280623365692
     },
     {
       "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": 9340.917840143406,
-      "required_pages_per_second_upper_estimate": 0.0037262317856005286
+      "required_token_positions_per_second": 9310.78306281312,
+      "required_pages_per_second_upper_estimate": 0.0037142105723684056
     },
     {
       "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": 18681.83568028681,
-      "required_pages_per_second_upper_estimate": 0.004321713895555922
+      "required_token_positions_per_second": 18621.56612562624,
+      "required_pages_per_second_upper_estimate": 0.004307771594793133
     },
     {
       "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": 28022.753520430215,
-      "required_pages_per_second_upper_estimate": 0.0017064386653690938
+      "required_token_positions_per_second": 27932.34918843936,
+      "required_pages_per_second_upper_estimate": 0.0017009335158657236
     },
     {
       "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": 28022.753520430215,
-      "required_pages_per_second_upper_estimate": 0.002157269137949288
+      "required_token_positions_per_second": 27932.34918843936,
+      "required_pages_per_second_upper_estimate": 0.002150309562217424
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28022.753520430215,
-      "required_pages_per_second_upper_estimate": 0.34794881260079485
+      "required_token_positions_per_second": 27932.34918843936,
+      "required_pages_per_second_upper_estimate": 0.3468262933613042
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07146317600779639,
-    "producer_p95_seconds": 0.18569748099980643,
-    "gpu_update_seconds": 0.9866268130946876,
-    "producer_batches_per_second": 13.993220786757412,
-    "gpu_batches_per_second": 1.0135544531405605,
-    "producer_headroom": 13.806086829768802,
+    "producer_batch_seconds": 0.0763146342578338,
+    "producer_p95_seconds": 0.1888923110000178,
+    "gpu_update_seconds": 0.9898200761231696,
+    "producer_batches_per_second": 13.103646629838217,
+    "gpu_batches_per_second": 1.0102846205309375,
+    "producer_headroom": 12.970252504637578,
     "sustainable": true,
-    "buffer_seconds": 60.18423559877594,
+    "buffer_seconds": 60.37902464351335,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311377.3636038103,
+    "generation_bytes_per_second": 330772.159238195,
     "upload_bytes_per_second": 5198660.677549234,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.81433373532994,
+    "estimated_full_shards_per_hour": 70.97600539073359,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -55,3 +55,4 @@
 {"step": 200500, "loss": 0.5637441445142031, "lr": 0.0003, "grad_norm": 0.15372752332988915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29345458656, "tokens_per_second": 148831.16892741164, "data_wait_s": 1.547899955767207e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 200600, "loss": 0.5990448854863644, "lr": 0.0003, "grad_norm": 0.15227823338963203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29360138716, "tokens_per_second": 149113.14160597263, "data_wait_s": 1.5094000445969868e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 200700, "loss": 0.5827857460826635, "lr": 0.0003, "grad_norm": 0.13504865827365398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29374811322, "tokens_per_second": 148735.56069760316, "data_wait_s": 1.6434999452030752e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 200800, "loss": 0.6274865567684174, "lr": 0.0003, "grad_norm": 0.1455793425619623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29389442497, "tokens_per_second": 148080.94717561323, "data_wait_s": 2.1547999494941905e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:00097600e4915da248a311b041be558f8f55692c4d4bc97bed1c433e7bb9e972
-size 121674
+oid sha256:b3f4ae6ab9ec4176b8283016fee344e72cfc1456df03fa2bd6b2763fdfd784ec
+size 123792