Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77549.3144797338,
-      "required_pages_per_second_upper_estimate": 0.005365041514908362
+      "required_token_positions_per_second": 76303.6419599075,
+      "required_pages_per_second_upper_estimate": 0.005278862999628318
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9693.664309966725,
-      "required_pages_per_second_upper_estimate": 0.004058534819853565
+      "required_token_positions_per_second": 9537.955244988438,
+      "required_pages_per_second_upper_estimate": 0.0039933426859221815
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19387.32861993345,
-      "required_pages_per_second_upper_estimate": 0.004037869273294458
+      "required_token_positions_per_second": 19075.910489976875,
+      "required_pages_per_second_upper_estimate": 0.003973009089472867
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29080.992929900174,
-      "required_pages_per_second_upper_estimate": 0.0017433611220423926
+      "required_token_positions_per_second": 28613.865734965315,
+      "required_pages_per_second_upper_estimate": 0.0017153575599679752
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29080.992929900174,
-      "required_pages_per_second_upper_estimate": 0.002260919286530184
+      "required_token_positions_per_second": 28613.865734965315,
+      "required_pages_per_second_upper_estimate": 0.002224602201799384
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29080.992929900174,
-      "required_pages_per_second_upper_estimate": 0.3730197525673115
+      "required_token_positions_per_second": 28613.865734965315,
+      "required_pages_per_second_upper_estimate": 0.3670279464727917
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0806372461407534,
-    "producer_p95_seconds": 0.29801099999895087,
-    "gpu_update_seconds": 0.9507240714457581,
-    "producer_batches_per_second": 12.401217152858699,
-    "gpu_batches_per_second": 1.0518298947446534,
-    "producer_headroom": 11.790135662448794,
+    "producer_batch_seconds": 0.4637542555310574,
+    "producer_p95_seconds": 0.27481902899671695,
+    "gpu_update_seconds": 0.9662448358459635,
+    "producer_batches_per_second": 2.156314444715711,
+    "gpu_batches_per_second": 1.034934379881558,
+    "producer_headroom": 2.083527696666612,
     "sustainable": true,
-    "buffer_seconds": 60.84634057252852,
+    "buffer_seconds": 60.873424658295704,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372147.6114637014,
-    "upload_bytes_per_second": 4880538.787092974,
+    "generation_bytes_per_second": 309844.9699834963,
+    "upload_bytes_per_second": 5009870.380950498,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.85421426709443,
+    "estimated_full_shards_per_hour": 66.48551773670832,
     "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": 0.9825653119987692,
-      "bytes": 2330980,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.2888210940000135,
       "bytes": 6181272,
@@ -562,6 +556,12 @@
       "bytes": 6332578,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.5906551469997794,
+      "bytes": 16693261,
+      "pages": 23,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -261,3 +261,4 @@
 {"step": 911100, "loss": 0.5160743594169617, "lr": 3e-05, "grad_norm": 0.3425965807331723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133358721328, "tokens_per_second": 152653.50272130972, "data_wait_s": 1.763399995979853e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 911200, "loss": 0.5397869944572449, "lr": 3e-05, "grad_norm": 0.31716942534774767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133373456651, "tokens_per_second": 153457.29424996313, "data_wait_s": 1.3385997590376064e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 911300, "loss": 0.5353308916091919, "lr": 3e-05, "grad_norm": 0.32864428744185187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133388201077, "tokens_per_second": 152774.58289776597, "data_wait_s": 1.0043000656878576e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 911400, "loss": 0.5455825924873352, "lr": 3e-05, "grad_norm": 0.32516339196001465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133403029935, "tokens_per_second": 154391.30851956684, "data_wait_s": 1.600700124981813e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71da476af0b8b37823aab41b6fc2ca55bc12f19d41c7c005cd60f917e499b03b
-size 561114
+oid sha256:376bfc677e56ed04ffa2b02f209d7fcda0a087523fe622be9ae3cd8e78fd4fd2
+size 563232