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": 76632.20794016913,
-      "required_pages_per_second_upper_estimate": 0.005301593956520927
+      "required_token_positions_per_second": 76502.93118848166,
+      "required_pages_per_second_upper_estimate": 0.005292650290875797
     },
     {
       "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": 9579.025992521141,
-      "required_pages_per_second_upper_estimate": 0.004010538150259389
+      "required_token_positions_per_second": 9562.866398560207,
+      "required_pages_per_second_upper_estimate": 0.004003772465718641
     },
     {
       "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": 19158.051985042282,
-      "required_pages_per_second_upper_estimate": 0.003990116996678094
+      "required_token_positions_per_second": 19125.732797120414,
+      "required_pages_per_second_upper_estimate": 0.003983385762148271
     },
     {
       "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": 28737.077977563422,
-      "required_pages_per_second_upper_estimate": 0.0017227439457775306
+      "required_token_positions_per_second": 28688.599195680625,
+      "required_pages_per_second_upper_estimate": 0.001719837716826471
     },
     {
       "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": 28737.077977563422,
-      "required_pages_per_second_upper_estimate": 0.002234181411708007
+      "required_token_positions_per_second": 28688.599195680625,
+      "required_pages_per_second_upper_estimate": 0.002230412399652246
     },
     {
       "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": 28737.077977563422,
-      "required_pages_per_second_upper_estimate": 0.3686083808258414
+      "required_token_positions_per_second": 28688.599195680625,
+      "required_pages_per_second_upper_estimate": 0.3679865470643094
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07040807353925516,
-    "producer_p95_seconds": 0.261438099998486,
-    "gpu_update_seconds": 0.9621019931666774,
-    "producer_batches_per_second": 14.202916650495519,
-    "gpu_batches_per_second": 1.0393908412023807,
-    "producer_headroom": 13.664654418221929,
+    "producer_batch_seconds": 0.8029755948437582,
+    "producer_p95_seconds": 0.2720080399994913,
+    "gpu_update_seconds": 0.963727779506317,
+    "producer_batches_per_second": 1.2453678622630846,
+    "gpu_batches_per_second": 1.0376374130382169,
+    "producer_headroom": 1.2001956045673314,
     "sustainable": true,
-    "buffer_seconds": 60.61242556950068,
+    "buffer_seconds": 60.71485010889798,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321152.27765402716,
-    "upload_bytes_per_second": 3550115.174798691,
+    "generation_bytes_per_second": 274855.675370583,
+    "upload_bytes_per_second": 3608555.4110745294,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.91180274215327,
+    "estimated_full_shards_per_hour": 58.977629621869255,
     "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": 1.6955216110000038,
-      "bytes": 7314599,
-      "pages": 30,
-      "shards": 3
-    },
-    {
-      "seconds": 1.2749757900000986,
-      "bytes": 5237846,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.5798416210000141,
       "bytes": 6237294,
@@ -562,6 +550,18 @@
       "bytes": 5900678,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8595360259969311,
+      "bytes": 1182793,
+      "pages": 35,
+      "shards": 1
+    },
+    {
+      "seconds": 1.7856543759990018,
+      "bytes": 17299146,
+      "pages": 27,
+      "shards": 6
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -205,3 +205,4 @@
 {"step": 863000, "loss": 0.5354397892951965, "lr": 3e-05, "grad_norm": 0.3094055649910224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126318981594, "tokens_per_second": 152018.03772973886, "data_wait_s": 9.193001460516825e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 863100, "loss": 0.49732932448387146, "lr": 3e-05, "grad_norm": 0.31005338994732745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126333450445, "tokens_per_second": 150386.20921807087, "data_wait_s": 1.0556999768596143e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 863200, "loss": 0.548261284828186, "lr": 3e-05, "grad_norm": 0.30352197081077575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126348016210, "tokens_per_second": 104951.52948531444, "data_wait_s": 2.045100336545147e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 863300, "loss": 0.53046053647995, "lr": 3e-05, "grad_norm": 0.3072887882862738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126362699196, "tokens_per_second": 152343.5660354792, "data_wait_s": 8.579001587349921e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91fc985df604a22a2ebcb95254b2f787a7c13be405c3ab30056ba928a83eb87b
-size 441723
+oid sha256:ccbdc2f984a62c6b153b63e6d51766f9934df68c05cf3c245ce07cf58f72b682
+size 443841