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": 75187.87582822153,
-      "required_pages_per_second_upper_estimate": 0.006517094514861157
+      "required_token_positions_per_second": 73864.32887166923,
+      "required_pages_per_second_upper_estimate": 0.006402372819166297
     },
     {
       "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": 9398.484478527691,
-      "required_pages_per_second_upper_estimate": 0.0037491959783499643
+      "required_token_positions_per_second": 9233.041108958654,
+      "required_pages_per_second_upper_estimate": 0.003683198144630068
     },
     {
       "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": 18796.968957055382,
-      "required_pages_per_second_upper_estimate": 0.0043483479528598655
+      "required_token_positions_per_second": 18466.08221791731,
+      "required_pages_per_second_upper_estimate": 0.004271803129167987
     },
     {
       "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": 28195.453435583073,
-      "required_pages_per_second_upper_estimate": 0.001716955184115474
+      "required_token_positions_per_second": 27699.123326875964,
+      "required_pages_per_second_upper_estimate": 0.0016867312845380404
     },
     {
       "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": 28195.453435583073,
-      "required_pages_per_second_upper_estimate": 0.0021705640554817176
+      "required_token_positions_per_second": 27699.123326875964,
+      "required_pages_per_second_upper_estimate": 0.002132355189783768
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28195.453435583073,
-      "required_pages_per_second_upper_estimate": 0.3500931675575583
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27699.123326875964,
+      "required_pages_per_second_upper_estimate": 0.3439304087174338
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08555045376556336,
-    "producer_p95_seconds": 0.3201156809955137,
-    "gpu_update_seconds": 0.9805836271853611,
-    "producer_batches_per_second": 11.689008719232886,
-    "gpu_batches_per_second": 1.0198008331735775,
-    "producer_headroom": 11.462050568106696,
+    "producer_batch_seconds": 0.43396939821906244,
+    "producer_p95_seconds": 0.3518187239969848,
+    "gpu_update_seconds": 0.9981543341183524,
+    "producer_batches_per_second": 2.3043099446731317,
+    "gpu_batches_per_second": 1.0018490786630485,
+    "producer_headroom": 2.3000569584275072,
     "sustainable": true,
-    "buffer_seconds": 60.79618488549239,
+    "buffer_seconds": 60.8874143812195,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390327.88351984415,
-    "upload_bytes_per_second": 4067954.1604864304,
+    "generation_bytes_per_second": 374995.0529162631,
+    "upload_bytes_per_second": 4151389.33518897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.75527743526929,
+    "estimated_full_shards_per_hour": 80.46520891776963,
     "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": 1.0622286870020616,
-      "bytes": 2500050,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.1363001400022767,
       "bytes": 9937212,
@@ -562,6 +556,12 @@
       "bytes": 6268819,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2165552439983003,
+      "bytes": 11720658,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 109800, "loss": 0.5832493174821138, "lr": 0.0003, "grad_norm": 0.16027330536996562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16067239572, "tokens_per_second": 148674.31861393442, "data_wait_s": 1.4754004951100796e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 109900, "loss": 0.6664679627865553, "lr": 0.0003, "grad_norm": 0.15876559523703643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16081894719, "tokens_per_second": 148556.2556977197, "data_wait_s": 1.6967998817563057e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 110000, "loss": 0.5629305355250835, "lr": 0.0003, "grad_norm": 0.13700414221724075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16096459430, "tokens_per_second": 147489.31812914196, "data_wait_s": 1.3880002370569855e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 110100, "loss": 0.5995736382901669, "lr": 0.0003, "grad_norm": 0.14184116113695153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16111112567, "tokens_per_second": 130778.05111079352, "data_wait_s": 2.0277999283280224e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa2600fdf19952bb924429cc61271211bf1d1f38bb5ceefff37a7df15620e8d4
-size 533580
+oid sha256:42ec4a9ed24e7f4463131b7a75ab2a1fb35c5864c439b0e667a7e0a8d433295f
+size 535698