Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74615.96217969531,
-      "required_pages_per_second_upper_estimate": 0.006467522489308796
+      "required_token_positions_per_second": 75068.58355314218,
+      "required_pages_per_second_upper_estimate": 0.006506754562800786
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9326.995272461914,
-      "required_pages_per_second_upper_estimate": 0.003720677865191445
+      "required_token_positions_per_second": 9383.572944142772,
+      "required_pages_per_second_upper_estimate": 0.0037432475443365137
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18653.99054492383,
-      "required_pages_per_second_upper_estimate": 0.004315272413547341
+      "required_token_positions_per_second": 18767.145888285544,
+      "required_pages_per_second_upper_estimate": 0.004341448910768047
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27980.985817385743,
-      "required_pages_per_second_upper_estimate": 0.0017038952313918874
+      "required_token_positions_per_second": 28150.718832428316,
+      "required_pages_per_second_upper_estimate": 0.0017142310814876733
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27980.985817385743,
-      "required_pages_per_second_upper_estimate": 0.0021540537445485226
+      "required_token_positions_per_second": 28150.718832428316,
+      "required_pages_per_second_upper_estimate": 0.0021671202618975573
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27980.985817385743,
-      "required_pages_per_second_upper_estimate": 0.34743019751649235
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28150.718832428316,
+      "required_pages_per_second_upper_estimate": 0.34953771350346197
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.15838886547652464,
-    "producer_p95_seconds": 0.25946448700051405,
-    "gpu_update_seconds": 0.9880995680581474,
-    "producer_batches_per_second": 6.313575117742184,
-    "gpu_batches_per_second": 1.012043757862621,
-    "producer_headroom": 6.2384408467437185,
+    "producer_batch_seconds": 0.24423327584378285,
+    "producer_p95_seconds": 0.23379406099957123,
+    "gpu_update_seconds": 0.9821418829330494,
+    "producer_batches_per_second": 4.094446166457771,
+    "gpu_batches_per_second": 1.0181828281404917,
+    "producer_headroom": 4.021327067492841,
     "sustainable": true,
-    "buffer_seconds": 60.27407365154699,
+    "buffer_seconds": 60.89279674184906,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400354.3176121502,
-    "upload_bytes_per_second": 5357125.110547315,
+    "generation_bytes_per_second": 380251.1586626084,
+    "upload_bytes_per_second": 5278821.496611803,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.90671678803807,
+    "estimated_full_shards_per_hour": 81.5930468550557,
     "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": 1.2473491540004034,
-      "bytes": 13161042,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.0770104249995711,
       "bytes": 6141587,
@@ -562,6 +556,12 @@
       "bytes": 6045685,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.297586668999429,
+      "bytes": 7207958,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -139,3 +139,4 @@
 {"step": 208900, "loss": 0.6065094582736492, "lr": 0.0003, "grad_norm": 0.172042730851505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30575317023, "tokens_per_second": 148050.1568811849, "data_wait_s": 1.7082998965634033e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209000, "loss": 0.6019245516508818, "lr": 0.0003, "grad_norm": 0.151780107319525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30589992924, "tokens_per_second": 147722.52947019503, "data_wait_s": 1.7903999832924455e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209100, "loss": 0.6039793342351913, "lr": 0.0003, "grad_norm": 0.1505738887339764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30604528995, "tokens_per_second": 146527.98548877737, "data_wait_s": 2.1677000404451974e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 209200, "loss": 0.5919628143310547, "lr": 0.0003, "grad_norm": 0.15929568507064515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30619022799, "tokens_per_second": 146032.3033093075, "data_wait_s": 1.7292000848101452e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37685a5220dbcc80cf995f454732e9fcb1a12f98255b2ae6c6a1ab42a45e68ed
-size 300369
+oid sha256:26ad6fa4b1d9255b1e5aaa678e83fc323e59989d0f5b9437b2d73599a3f6a716
+size 302487