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": 76672.9395873304,
-      "required_pages_per_second_upper_estimate": 0.005304411866381986
+      "required_token_positions_per_second": 76675.13579056322,
+      "required_pages_per_second_upper_estimate": 0.005304563804817526
     },
     {
       "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": 9584.1174484163,
-      "required_pages_per_second_upper_estimate": 0.004012669836521003
+      "required_token_positions_per_second": 9584.391973820402,
+      "required_pages_per_second_upper_estimate": 0.004012784774574958
     },
     {
       "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": 19168.2348968326,
-      "required_pages_per_second_upper_estimate": 0.003992237828662575
+      "required_token_positions_per_second": 19168.783947640804,
+      "required_pages_per_second_upper_estimate": 0.003992352181466481
     },
     {
       "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": 28752.3523452489,
-      "required_pages_per_second_upper_estimate": 0.0017236596207976659
+      "required_token_positions_per_second": 28753.175921461207,
+      "required_pages_per_second_upper_estimate": 0.0017237089929340657
     },
     {
       "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": 28752.3523452489,
-      "required_pages_per_second_upper_estimate": 0.0022353689266107098
+      "required_token_positions_per_second": 28753.175921461207,
+      "required_pages_per_second_upper_estimate": 0.0022354329560386878
     },
     {
       "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": 28752.3523452489,
-      "required_pages_per_second_upper_estimate": 0.36880430401417247
+      "required_token_positions_per_second": 28753.175921461207,
+      "required_pages_per_second_upper_estimate": 0.3688148679655367
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5456554794218818,
-    "producer_p95_seconds": 0.18641234499955317,
-    "gpu_update_seconds": 0.9615908871737451,
-    "producer_batches_per_second": 1.832658220640418,
-    "gpu_batches_per_second": 1.039943299524338,
-    "producer_headroom": 1.7622674442718766,
+    "producer_batch_seconds": 0.07475280573443399,
+    "producer_p95_seconds": 0.18872244699923613,
+    "gpu_update_seconds": 0.9615633443595945,
+    "producer_batches_per_second": 13.377424301003352,
+    "gpu_batches_per_second": 1.0399730874371096,
+    "producer_headroom": 12.863240849790094,
     "sustainable": true,
-    "buffer_seconds": 60.58022589194594,
+    "buffer_seconds": 60.57849069465445,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299772.8215556846,
+    "generation_bytes_per_second": 287294.3630846701,
     "upload_bytes_per_second": 4828335.407220306,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.32426915171531,
+    "estimated_full_shards_per_hour": 61.64668244748189,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -92,3 +92,4 @@
 {"step": 729200, "loss": 0.5399318337440491, "lr": 5e-05, "grad_norm": 0.215438854976121, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106734428913, "tokens_per_second": 151357.1710056955, "data_wait_s": 1.3409000530373305e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 729300, "loss": 0.5082528591156006, "lr": 5e-05, "grad_norm": 0.21188878347115842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106749031340, "tokens_per_second": 141508.57364701325, "data_wait_s": 9.015999239636585e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 729400, "loss": 0.5691736936569214, "lr": 5e-05, "grad_norm": 0.23089856002698342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106763644236, "tokens_per_second": 151588.10487979575, "data_wait_s": 1.0775000191642903e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 729500, "loss": 0.5539270639419556, "lr": 5e-05, "grad_norm": 0.22316113871322815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106778141226, "tokens_per_second": 149455.07718477384, "data_wait_s": 1.3678998584509827e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:914fd61204e9e916401f0216262c19744861a6023b078d6429d37a9956fbd1dd
-size 200040
+oid sha256:2fb3e2fee3b1e68a4465826c938475bbb6390daec144619fa31084ee6fad2c34
+size 202158