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": 74940.12533214892,
-      "required_pages_per_second_upper_estimate": 0.006495620129779475
+      "required_token_positions_per_second": 75081.92265198388,
+      "required_pages_per_second_upper_estimate": 0.006507910762080749
     },
     {
       "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": 9367.515666518615,
-      "required_pages_per_second_upper_estimate": 0.0037368420562145423
+      "required_token_positions_per_second": 9385.240331497986,
+      "required_pages_per_second_upper_estimate": 0.003743912690082171
     },
     {
       "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": 18735.03133303723,
-      "required_pages_per_second_upper_estimate": 0.004334019773608219
+      "required_token_positions_per_second": 18770.48066299597,
+      "required_pages_per_second_upper_estimate": 0.0043422203522072746
     },
     {
       "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": 28102.546999555845,
-      "required_pages_per_second_upper_estimate": 0.001711297669603814
+      "required_token_positions_per_second": 28155.720994493957,
+      "required_pages_per_second_upper_estimate": 0.0017145356869131547
     },
     {
       "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": 28102.546999555845,
-      "required_pages_per_second_upper_estimate": 0.002163411860854866
+      "required_token_positions_per_second": 28155.720994493957,
+      "required_pages_per_second_upper_estimate": 0.002167505342180241
     },
     {
       "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": 28102.546999555845,
-      "required_pages_per_second_upper_estimate": 0.34893958055993946
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28155.720994493957,
+      "required_pages_per_second_upper_estimate": 0.34959982361515773
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.42891186915627344,
-    "producer_p95_seconds": 0.17933583200010617,
-    "gpu_update_seconds": 0.9838254162669657,
-    "producer_batches_per_second": 2.331481294670471,
-    "gpu_batches_per_second": 1.0164405020093985,
-    "producer_headroom": 2.29377055524782,
+    "producer_batch_seconds": 0.05298155201560917,
+    "producer_p95_seconds": 0.17326564500035602,
+    "gpu_update_seconds": 0.9819673950245051,
+    "producer_batches_per_second": 18.874494271239634,
+    "gpu_batches_per_second": 1.0183637512476114,
+    "producer_headroom": 18.53413797193413,
     "sustainable": true,
-    "buffer_seconds": 60.01335039228491,
+    "buffer_seconds": 60.88197849151932,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356382.69081817265,
-    "upload_bytes_per_second": 5393018.259402642,
+    "generation_bytes_per_second": 356264.2300616495,
+    "upload_bytes_per_second": 5563706.282868056,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.47142928513416,
+    "estimated_full_shards_per_hour": 76.44601036440959,
     "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": [
@@ -262,6 +262,12 @@
       "bytes": 6754320,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.6214569329995356,
+      "bytes": 11854578,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -46,3 +46,4 @@
 {"step": 199600, "loss": 0.5836126524955034, "lr": 0.0003, "grad_norm": 0.14907554180658394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29213631440, "tokens_per_second": 149787.08687137134, "data_wait_s": 2.0182999833195936e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199700, "loss": 0.5929410867393017, "lr": 0.0003, "grad_norm": 0.14810363831661824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29228097396, "tokens_per_second": 145247.3200887589, "data_wait_s": 1.8230999557999894e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199800, "loss": 0.5837657302618027, "lr": 0.0003, "grad_norm": 0.13412700037772263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29242595259, "tokens_per_second": 146476.28526405295, "data_wait_s": 1.7119000403909013e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 199900, "loss": 0.5994546860456467, "lr": 0.0003, "grad_norm": 0.1577259501370254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29257211636, "tokens_per_second": 147569.86863726124, "data_wait_s": 1.5909000467217993e-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:c936527e225ce6addd4a3fb93f3ab1cbf4b5dff2883593331b562149dec54d66
-size 101829
+oid sha256:bbaccb77fd4b28bbec290c6961bbf59d7f3bd0a1ced68f7990f11d5c2d8f3987
+size 103947