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": 75278.53433249312,
-      "required_pages_per_second_upper_estimate": 0.005207943675386391
+      "required_token_positions_per_second": 75757.38632189925,
+      "required_pages_per_second_upper_estimate": 0.005241071767103201
     },
     {
       "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": 9409.81679156164,
-      "required_pages_per_second_upper_estimate": 0.003939693791307568
+      "required_token_positions_per_second": 9469.673290237406,
+      "required_pages_per_second_upper_estimate": 0.003964754457357283
     },
     {
       "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": 18819.63358312328,
-      "required_pages_per_second_upper_estimate": 0.003919633368251785
+      "required_token_positions_per_second": 18939.346580474812,
+      "required_pages_per_second_upper_estimate": 0.0039445664285560665
     },
     {
       "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": 28229.450374684922,
-      "required_pages_per_second_upper_estimate": 0.0016923124460874342
+      "required_token_positions_per_second": 28409.019870712218,
+      "required_pages_per_second_upper_estimate": 0.0017030773631875262
     },
     {
       "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": 28229.450374684922,
-      "required_pages_per_second_upper_estimate": 0.0021947155984020574
+      "required_token_positions_per_second": 28409.019870712218,
+      "required_pages_per_second_upper_estimate": 0.00220867633687544
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28229.450374684922,
-      "required_pages_per_second_upper_estimate": 0.3620970789841706
+      "required_token_positions_per_second": 28409.019870712218,
+      "required_pages_per_second_upper_estimate": 0.3644004036725057
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.17110320267192947,
-    "producer_p95_seconds": 0.2710215100014466,
-    "gpu_update_seconds": 0.9794027029585265,
-    "producer_batches_per_second": 5.844425962717857,
-    "gpu_batches_per_second": 1.0210304678343793,
-    "producer_headroom": 5.724046585126858,
+    "producer_batch_seconds": 0.0926762325702839,
+    "producer_p95_seconds": 0.2574918859972968,
+    "gpu_update_seconds": 0.9732120335662555,
+    "producer_batches_per_second": 10.790253037547885,
+    "gpu_batches_per_second": 1.0275253136108298,
+    "producer_headroom": 10.501204101366442,
     "sustainable": true,
-    "buffer_seconds": 60.72296758342864,
+    "buffer_seconds": 60.33914608110784,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263577.2742741723,
-    "upload_bytes_per_second": 3921184.8852883764,
+    "generation_bytes_per_second": 294707.63637496385,
+    "upload_bytes_per_second": 3919596.846381597,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.55754729431989,
+    "estimated_full_shards_per_hour": 63.237398323408954,
     "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": 1.5393589430004795,
-      "bytes": 11424070,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.9194490070003667,
       "bytes": 6288888,
@@ -562,6 +556,12 @@
       "bytes": 917525,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 2.0703852710030333,
+      "bytes": 13336105,
+      "pages": 42,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -210,3 +210,4 @@
 {"step": 821000, "loss": 0.5010907053947449, "lr": 3e-05, "grad_norm": 0.2722542619171817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120171463232, "tokens_per_second": 151998.30985808626, "data_wait_s": 1.1166997865075245e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821100, "loss": 0.5874857306480408, "lr": 3e-05, "grad_norm": 0.29996884821400993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120186205649, "tokens_per_second": 152240.61913553393, "data_wait_s": 1.7594000382814556e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821200, "loss": 0.5458189249038696, "lr": 3e-05, "grad_norm": 0.2782306516437237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120200872674, "tokens_per_second": 151976.70244240065, "data_wait_s": 1.0406998626422137e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 821300, "loss": 0.48540323972702026, "lr": 3e-05, "grad_norm": 0.2919788006193103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120215390016, "tokens_per_second": 149984.20810734577, "data_wait_s": 1.160899773822166e-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 sha256:272fb10564d9dab3d9198f4b225e6a8e49ea5e618fe8221899191b4fe827de2b
-size 452313
+oid sha256:aaad72022ab4b89d4ec9561a02ff16845a7a11da0aeeda0bc7ac881d8c3dcb4a
+size 454431