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": 73518.76759854486,
-      "required_pages_per_second_upper_estimate": 0.0063724204438289165
+      "required_token_positions_per_second": 75184.2515735437,
+      "required_pages_per_second_upper_estimate": 0.006516780373651273
     },
     {
       "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": 9189.845949818107,
-      "required_pages_per_second_upper_estimate": 0.0036659669498237224
+      "required_token_positions_per_second": 9398.031446692963,
+      "required_pages_per_second_upper_estimate": 0.003749015257177662
     },
     {
       "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": 18379.691899636215,
-      "required_pages_per_second_upper_estimate": 0.0042518182494513295
+      "required_token_positions_per_second": 18796.062893385926,
+      "required_pages_per_second_upper_estimate": 0.004348138350976113
     },
     {
       "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": 27569.537849454322,
-      "required_pages_per_second_upper_estimate": 0.0016788402088455184
+      "required_token_positions_per_second": 28194.09434007889,
+      "required_pages_per_second_upper_estimate": 0.0017168724223298923
     },
     {
       "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": 27569.537849454322,
-      "required_pages_per_second_upper_estimate": 0.0021223793410162896
+      "required_token_positions_per_second": 28194.09434007889,
+      "required_pages_per_second_upper_estimate": 0.0021704594285475885
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27569.537849454322,
-      "required_pages_per_second_upper_estimate": 0.34232139078255114
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28194.09434007889,
+      "required_pages_per_second_upper_estimate": 0.3500762921399964
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11943556728925842,
-    "producer_p95_seconds": 0.394949263005401,
-    "gpu_update_seconds": 1.0028459726446677,
-    "producer_batches_per_second": 8.372715286545436,
-    "gpu_batches_per_second": 0.9971621039299162,
-    "producer_headroom": 8.396543805212536,
+    "producer_batch_seconds": 0.15105090650774855,
+    "producer_p95_seconds": 0.39483053099684184,
+    "gpu_update_seconds": 0.9806308961908169,
+    "producer_batches_per_second": 6.620284665082115,
+    "gpu_batches_per_second": 1.0197516760734553,
+    "producer_headroom": 6.492055684157797,
     "sustainable": true,
-    "buffer_seconds": 60.17075835868006,
+    "buffer_seconds": 60.79911556383065,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 494372.385505368,
-    "upload_bytes_per_second": 5820386.720231007,
+    "generation_bytes_per_second": 509092.5437669178,
+    "upload_bytes_per_second": 5798216.939113974,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.08080552931575,
+    "estimated_full_shards_per_hour": 109.2394088244977,
     "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": 1.031183188999421,
-      "bytes": 7489997,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.171489565000229,
       "bytes": 5465773,
@@ -562,6 +556,12 @@
       "bytes": 5894461,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3212683820020175,
+      "bytes": 7578218,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -342,3 +342,4 @@
 {"step": 229200, "loss": 0.650917025282979, "lr": 0.0003, "grad_norm": 0.1509130138788284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33548146126, "tokens_per_second": 149245.12045595492, "data_wait_s": 1.3757002307102084e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 229300, "loss": 0.5776455719023943, "lr": 0.0003, "grad_norm": 0.15469473841681558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33562895654, "tokens_per_second": 150559.20578385357, "data_wait_s": 1.35790032800287e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 229400, "loss": 0.6375050283968449, "lr": 0.0003, "grad_norm": 0.15609010654636596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33577628688, "tokens_per_second": 149396.0734458478, "data_wait_s": 1.832599809858948e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 229500, "loss": 0.5922210942953825, "lr": 0.0003, "grad_norm": 0.12819053892824425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33592338050, "tokens_per_second": 149916.34407644212, "data_wait_s": 1.1964999430347234e-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 sha256:553902217b6714ce5cee564eca5c83ee58d32be6815804457d2c3c49f405072a
-size 733455
+oid sha256:3bfe3cb62fe53c3fcd1a49fb3932ab32c2f20feed0208a8e1cea11dd765388c9
+size 735573