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": 75293.24548794466,
-      "required_pages_per_second_upper_estimate": 0.006526227689908997
+      "required_token_positions_per_second": 75144.8304475347,
+      "required_pages_per_second_upper_estimate": 0.006513363450360743
     },
     {
       "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": 9411.655685993082,
-      "required_pages_per_second_upper_estimate": 0.00375445016993501
+      "required_token_positions_per_second": 9393.103805941837,
+      "required_pages_per_second_upper_estimate": 0.003747049547607243
     },
     {
       "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": 18823.311371986165,
-      "required_pages_per_second_upper_estimate": 0.004354441806943434
+      "required_token_positions_per_second": 18786.207611883674,
+      "required_pages_per_second_upper_estimate": 0.004345858505047598
     },
     {
       "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": 28234.967057979247,
-      "required_pages_per_second_upper_estimate": 0.0017193613564074463
+      "required_token_positions_per_second": 28179.31141782551,
+      "required_pages_per_second_upper_estimate": 0.0017159722199246604
     },
     {
       "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": 28234.967057979247,
-      "required_pages_per_second_upper_estimate": 0.0021736059235144827
+      "required_token_positions_per_second": 28179.31141782551,
+      "required_pages_per_second_upper_estimate": 0.0021693213982707754
     },
     {
       "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": 28234.967057979247,
-      "required_pages_per_second_upper_estimate": 0.35058379450413163
+      "required_token_positions_per_second": 28179.31141782551,
+      "required_pages_per_second_upper_estimate": 0.34989273772086754
     }
   ],
   "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": 0.3498680634526181,
-    "producer_p95_seconds": 0.4446488409957965,
-    "gpu_update_seconds": 0.9792113425606647,
-    "producer_batches_per_second": 2.8582202963358725,
-    "gpu_batches_per_second": 1.021230000650291,
-    "producer_headroom": 2.7988017337091904,
+    "producer_batch_seconds": 0.4514699378515843,
+    "producer_p95_seconds": 0.42196200900070835,
+    "gpu_update_seconds": 0.9811453370897695,
+    "producer_batches_per_second": 2.2149869042415373,
+    "gpu_batches_per_second": 1.0192169928322306,
+    "producer_headroom": 2.173224072811488,
     "sustainable": true,
-    "buffer_seconds": 60.71110323876121,
+    "buffer_seconds": 60.831010899565705,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374505.8306879986,
-    "upload_bytes_per_second": 3823029.6725944765,
+    "generation_bytes_per_second": 339100.96171123907,
+    "upload_bytes_per_second": 3871535.0426104316,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.36023321609467,
+    "estimated_full_shards_per_hour": 72.76317251685028,
     "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": 1.0071898829992278,
-      "bytes": 5466457,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1624265700011165,
       "bytes": 7895099,
@@ -562,6 +556,12 @@
       "bytes": 7618244,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1348141839989694,
+      "bytes": 11320691,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -384,3 +384,4 @@
 {"step": 353400, "loss": 0.6058970261365175, "lr": 0.0003, "grad_norm": 0.16720838608454577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51727941253, "tokens_per_second": 146409.04667974575, "data_wait_s": 1.5572994016110897e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 353500, "loss": 0.5445586293935776, "lr": 0.0003, "grad_norm": 0.16862535399243653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51742517664, "tokens_per_second": 147934.22499500206, "data_wait_s": 7.714996172580868e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 353600, "loss": 0.5894202198833227, "lr": 0.0003, "grad_norm": 0.1366230051877542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51756950256, "tokens_per_second": 134158.24155465205, "data_wait_s": 9.75099828792736e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 353700, "loss": 0.5857976600527763, "lr": 0.0003, "grad_norm": 0.15603513816315603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51771622082, "tokens_per_second": 148353.41420498275, "data_wait_s": 1.4192999515216798e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:56f6a032d970550bc3adc92e220b5ae07d06b90bc5a3b7c553f7fe06f67b1c1c
-size 823194
+oid sha256:a0873a8b003074012b7c973eb4f6021e54d078c62df80d129e982c8bb3828522
+size 825312