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": 72971.33236825085,
-      "required_pages_per_second_upper_estimate": 0.006324970145528938
+      "required_token_positions_per_second": 74416.08591476134,
+      "required_pages_per_second_upper_estimate": 0.006450197721246081
     },
     {
       "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": 9121.416546031356,
-      "required_pages_per_second_upper_estimate": 0.003638669437542427
+      "required_token_positions_per_second": 9302.010739345167,
+      "required_pages_per_second_upper_estimate": 0.0037107111613791156
     },
     {
       "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": 18242.833092062712,
-      "required_pages_per_second_upper_estimate": 0.004220158372951948
+      "required_token_positions_per_second": 18604.021478690334,
+      "required_pages_per_second_upper_estimate": 0.004303712949499752
     },
     {
       "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": 27364.24963809407,
-      "required_pages_per_second_upper_estimate": 0.001666339233837684
+      "required_token_positions_per_second": 27906.0322180355,
+      "required_pages_per_second_upper_estimate": 0.0016993309504426013
     },
     {
       "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": 27364.24963809407,
-      "required_pages_per_second_upper_estimate": 0.0021065756862316373
+      "required_token_positions_per_second": 27906.0322180355,
+      "required_pages_per_second_upper_estimate": 0.002148283608985698
     },
     {
       "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": 27364.24963809407,
-      "required_pages_per_second_upper_estimate": 0.33977239825290323
+      "required_token_positions_per_second": 27906.0322180355,
+      "required_pages_per_second_upper_estimate": 0.3464995246661224
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.264124182359609,
-    "producer_p95_seconds": 0.4007862250000471,
-    "gpu_update_seconds": 1.0103693821558666,
-    "producer_batches_per_second": 3.786097853919658,
-    "gpu_batches_per_second": 0.9897370384148607,
-    "producer_headroom": 3.8253573494464574,
+    "producer_batch_seconds": 0.12573980463275802,
+    "producer_p95_seconds": 0.3740923639998073,
+    "gpu_update_seconds": 0.9907535325688925,
+    "producer_batches_per_second": 7.952931077956182,
+    "gpu_batches_per_second": 1.0093327625157515,
+    "producer_headroom": 7.879394559762018,
     "sustainable": true,
-    "buffer_seconds": 60.622162929351994,
+    "buffer_seconds": 60.43596548670244,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364264.3610083762,
-    "upload_bytes_per_second": 3211899.764609378,
+    "generation_bytes_per_second": 364407.90253521147,
+    "upload_bytes_per_second": 3294400.5504429345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.16265223206008,
+    "estimated_full_shards_per_hour": 78.19345290224321,
     "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": 0.8964433900000586,
-      "bytes": 1180194,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.5248420139978407,
       "bytes": 7676145,
@@ -562,6 +556,12 @@
       "bytes": 999325,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.4190698960010195,
+      "bytes": 13694673,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -317,3 +317,4 @@
 {"step": 346700, "loss": 0.6234288327395916, "lr": 0.0003, "grad_norm": 0.1694497640742114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50747791494, "tokens_per_second": 138618.26169335644, "data_wait_s": 1.367900040349923e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346800, "loss": 0.5871155746281147, "lr": 0.0003, "grad_norm": 0.14986730547354005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50762460943, "tokens_per_second": 148582.02651873155, "data_wait_s": 1.7239999579032883e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346900, "loss": 0.5453714616596699, "lr": 0.0003, "grad_norm": 0.15822205503828132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50777211061, "tokens_per_second": 150061.0585118322, "data_wait_s": 1.5470999642275274e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 347000, "loss": 0.6010876540094614, "lr": 0.0003, "grad_norm": 0.15745685106639337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50792017038, "tokens_per_second": 150406.42502960155, "data_wait_s": 1.6705998859833926e-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:e5e6aa4d41f23ff92f5e3ba38f02a9823ed4f4093b9645b8ebb4fedea72fd4cd
-size 680505
+oid sha256:62f776b11bfe52c78c65d565a8780780492625176f3e78cfc3033edd3cefc27e
+size 682623