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": 76453.99465295758,
-      "required_pages_per_second_upper_estimate": 0.005289264747799835
+      "required_token_positions_per_second": 75083.69598763426,
+      "required_pages_per_second_upper_estimate": 0.00519446430660181
     },
     {
       "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": 9556.749331619698,
-      "required_pages_per_second_upper_estimate": 0.004001211377529533
+      "required_token_positions_per_second": 9385.461998454282,
+      "required_pages_per_second_upper_estimate": 0.003929496948019431
     },
     {
       "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": 19113.498663239396,
-      "required_pages_per_second_upper_estimate": 0.003980837714696661
+      "required_token_positions_per_second": 18770.923996908565,
+      "required_pages_per_second_upper_estimate": 0.00390948844600143
     },
     {
       "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": 28670.247994859095,
-      "required_pages_per_second_upper_estimate": 0.0017187375903578814
+      "required_token_positions_per_second": 28156.385995362845,
+      "required_pages_per_second_upper_estimate": 0.0016879323481099233
     },
     {
       "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": 28670.247994859095,
-      "required_pages_per_second_upper_estimate": 0.002228985674506774
+      "required_token_positions_per_second": 28156.385995362845,
+      "required_pages_per_second_upper_estimate": 0.0021890351642860064
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28670.247994859095,
-      "required_pages_per_second_upper_estimate": 0.3677511575641551
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28156.385995362845,
+      "required_pages_per_second_upper_estimate": 0.3611598875766453
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08216051328129481,
-    "producer_p95_seconds": 0.22235233899846207,
-    "gpu_update_seconds": 0.9643446406517866,
-    "producer_batches_per_second": 12.171296892660314,
-    "gpu_batches_per_second": 1.0369736687955402,
-    "producer_headroom": 11.737324928218717,
+    "producer_batch_seconds": 0.33192726177350096,
+    "producer_p95_seconds": 0.24335567799971614,
+    "gpu_update_seconds": 0.9819442028019302,
+    "producer_batches_per_second": 3.012708250165892,
+    "gpu_batches_per_second": 1.0183878036517233,
+    "producer_headroom": 2.958311400983945,
     "sustainable": true,
-    "buffer_seconds": 60.753712361062554,
+    "buffer_seconds": 60.88054057371967,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448699.9167768318,
+    "generation_bytes_per_second": 414260.947338046,
     "upload_bytes_per_second": 3901727.864121589,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.28055694082943,
+    "estimated_full_shards_per_hour": 88.89075579744372,
     "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": [

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 814900, "loss": 0.5868908166885376, "lr": 3e-05, "grad_norm": 0.2968352698170268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119278161782, "tokens_per_second": 149856.9553760005, "data_wait_s": 2.10089983738726e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 815000, "loss": 0.513067364692688, "lr": 3e-05, "grad_norm": 0.2654177458091664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119292759153, "tokens_per_second": 151537.77438656858, "data_wait_s": 1.0326000847271644e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 815100, "loss": 0.5704906582832336, "lr": 3e-05, "grad_norm": 0.28742790877356944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119307411443, "tokens_per_second": 137324.23337128272, "data_wait_s": 1.141199936682824e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 815200, "loss": 0.48602429032325745, "lr": 3e-05, "grad_norm": 0.280646224181739, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119322020275, "tokens_per_second": 151678.57879253337, "data_wait_s": 1.0450999980093911e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa4f252b063bc2ab9c89addf43386e804888f89fb659761816820d4986c7a8d1
-size 322332
+oid sha256:39f2c8e7968c0013618e22a4dbaa9449dbaf045cbfb06316bb6f796f7c9b5779
+size 324450