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": 77376.32192426031,
-      "required_pages_per_second_upper_estimate": 0.005353073488522674
+      "required_token_positions_per_second": 77272.2466294703,
+      "required_pages_per_second_upper_estimate": 0.005345873318141147
     },
     {
       "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": 9672.040240532539,
-      "required_pages_per_second_upper_estimate": 0.00404948127354339
+      "required_token_positions_per_second": 9659.030828683788,
+      "required_pages_per_second_upper_estimate": 0.004044034504469729
     },
     {
       "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": 19344.080481065077,
-      "required_pages_per_second_upper_estimate": 0.004028861826498295
+      "required_token_positions_per_second": 19318.061657367576,
+      "required_pages_per_second_upper_estimate": 0.004023442791684629
     },
     {
       "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": 29016.120721597617,
-      "required_pages_per_second_upper_estimate": 0.001739472132208781
+      "required_token_positions_per_second": 28977.092486051366,
+      "required_pages_per_second_upper_estimate": 0.0017371324490804483
     },
     {
       "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": 29016.120721597617,
-      "required_pages_per_second_upper_estimate": 0.0022558757576773484
+      "required_token_positions_per_second": 28977.092486051366,
+      "required_pages_per_second_upper_estimate": 0.0022528414840306933
     },
     {
       "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": 29016.120721597617,
-      "required_pages_per_second_upper_estimate": 0.37218764153355677
+      "required_token_positions_per_second": 28977.092486051366,
+      "required_pages_per_second_upper_estimate": 0.3716870292332239
     }
   ],
   "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": 0.050692104078081,
-    "producer_p95_seconds": 0.16794688799927826,
-    "gpu_update_seconds": 0.9528496336665955,
-    "producer_batches_per_second": 19.726938113669554,
-    "gpu_batches_per_second": 1.0494835330438954,
-    "producer_headroom": 18.796805754973636,
+    "producer_batch_seconds": 0.09272778437503604,
+    "producer_p95_seconds": 0.18866425699980027,
+    "gpu_update_seconds": 0.9541329936158658,
+    "producer_batches_per_second": 10.784254220455823,
+    "gpu_batches_per_second": 1.048071921515168,
+    "producer_headroom": 10.28961276327805,
     "sustainable": true,
-    "buffer_seconds": 60.029526920995515,
+    "buffer_seconds": 60.110378597799546,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 163,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 497218.48989701486,
-    "upload_bytes_per_second": 2673729.776303623,
+    "generation_bytes_per_second": 636769.8726239484,
+    "upload_bytes_per_second": 2626591.675765,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.69151327784381,
+    "estimated_full_shards_per_hour": 68.31799571055812,
     "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": [
@@ -292,6 +292,12 @@
       "bytes": 10306218,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.8383978550000393,
+      "bytes": 2628306,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -64,3 +64,4 @@
 {"step": 691400, "loss": 0.5636149644851685, "lr": 8e-05, "grad_norm": 0.20252056343074062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101202534590, "tokens_per_second": 152856.4707323732, "data_wait_s": 1.3744000170845538e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 691500, "loss": 0.6558876037597656, "lr": 8e-05, "grad_norm": 0.22496898712752755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101217243178, "tokens_per_second": 153730.30006375152, "data_wait_s": 1.851599972724216e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 691600, "loss": 0.5512977838516235, "lr": 8e-05, "grad_norm": 0.20203488715218076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101231965499, "tokens_per_second": 153709.78064743686, "data_wait_s": 1.5434999113494996e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 691700, "loss": 0.5753223299980164, "lr": 8e-05, "grad_norm": 0.2086336161622996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101246670272, "tokens_per_second": 153526.06875885683, "data_wait_s": 1.4373000340128783e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d92831615a0cc5e818b07f2333b579418c2f1b0f0355582c3cc12a932844de31
-size 140736
+oid sha256:1e220e800cf65529674a1f877f56a6762ef91049d2798c9918b6bedce2e2b261
+size 142854