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": 75616.35871084781,
-      "required_pages_per_second_upper_estimate": 0.0052313151497415215
+      "required_token_positions_per_second": 77377.1247397013,
+      "required_pages_per_second_upper_estimate": 0.005353129029157671
     },
     {
       "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": 9452.044838855976,
-      "required_pages_per_second_upper_estimate": 0.003957373792887804
+      "required_token_positions_per_second": 9672.140592462663,
+      "required_pages_per_second_upper_estimate": 0.004049523288800946
     },
     {
       "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": 18904.089677711952,
-      "required_pages_per_second_upper_estimate": 0.003937223345497614
+      "required_token_positions_per_second": 19344.281184925327,
+      "required_pages_per_second_upper_estimate": 0.004028903627819468
     },
     {
       "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": 28356.13451656793,
-      "required_pages_per_second_upper_estimate": 0.0016999069669578362
+      "required_token_positions_per_second": 29016.42177738799,
+      "required_pages_per_second_upper_estimate": 0.0017394901800437282
     },
     {
       "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": 28356.13451656793,
-      "required_pages_per_second_upper_estimate": 0.0022045647332088105
+      "required_token_positions_per_second": 29016.42177738799,
+      "required_pages_per_second_upper_estimate": 0.0022558991634408455
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28356.13451656793,
-      "required_pages_per_second_upper_estimate": 0.36372204713341194
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29016.42177738799,
+      "required_pages_per_second_upper_estimate": 0.37219150315398714
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08206898542192675,
-    "producer_p95_seconds": 0.2244220590000623,
-    "gpu_update_seconds": 0.9750271139335236,
-    "producer_batches_per_second": 12.184870994308957,
-    "gpu_batches_per_second": 1.0256125042161433,
-    "producer_headroom": 11.880579599233366,
+    "producer_batch_seconds": 0.18940992239849663,
+    "producer_p95_seconds": 0.24691206900024554,
+    "gpu_update_seconds": 0.9528397475096541,
+    "producer_batches_per_second": 5.279554457005243,
+    "gpu_batches_per_second": 1.0494944219252023,
+    "producer_headroom": 5.030569335776344,
     "sustainable": true,
-    "buffer_seconds": 60.45168106387846,
+    "buffer_seconds": 60.0289040931082,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 501959.7831191211,
-    "upload_bytes_per_second": 3410992.5144012915,
+    "generation_bytes_per_second": 497766.046612844,
+    "upload_bytes_per_second": 3461601.219718445,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.7088844316504,
+    "estimated_full_shards_per_hour": 106.80900620259276,
     "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": 1.021054885999547,
-      "bytes": 1163402,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.532974641999317,
       "bytes": 11431908,
@@ -562,6 +556,12 @@
       "bytes": 6147166,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.2317090710002958,
+      "bytes": 8202394,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -160,3 +160,4 @@
 {"step": 858500, "loss": 0.49152708053588867, "lr": 3e-05, "grad_norm": 0.284207025313917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125660434830, "tokens_per_second": 151467.72231918594, "data_wait_s": 1.0134999683941714e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858600, "loss": 0.5888546109199524, "lr": 3e-05, "grad_norm": 0.33249903731800506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125675099357, "tokens_per_second": 152094.5574051788, "data_wait_s": 1.5794999853824265e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858700, "loss": 0.576016902923584, "lr": 3e-05, "grad_norm": 0.3270641719528925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125689834326, "tokens_per_second": 152887.49473731048, "data_wait_s": 1.4182998711476102e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 858800, "loss": 0.498494952917099, "lr": 3e-05, "grad_norm": 0.29280652892263864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125704548608, "tokens_per_second": 150410.5881663116, "data_wait_s": 1.1326999810989946e-05, "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 sha256:7aee026f461524537441419e07eb903fb41f2dfc61f5008eb5931a617131c1f4
-size 345630
+oid sha256:1f2c87a4abea8be4a5f287e3d31cea9bf309921b9e507177eadce4125bc715a1
+size 347748