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": 74648.80171130507,
-      "required_pages_per_second_upper_estimate": 0.006470368936677697
+      "required_token_positions_per_second": 74649.35466884455,
+      "required_pages_per_second_upper_estimate": 0.006470416865635761
     },
     {
       "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": 9331.100213913134,
-      "required_pages_per_second_upper_estimate": 0.0037223153877106807
+      "required_token_positions_per_second": 9331.169333605569,
+      "required_pages_per_second_upper_estimate": 0.0037223429605894244
     },
     {
       "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": 18662.20042782627,
-      "required_pages_per_second_upper_estimate": 0.004317171625467816
+      "required_token_positions_per_second": 18662.338667211137,
+      "required_pages_per_second_upper_estimate": 0.004317203604716576
     },
     {
       "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": 27993.300641739403,
-      "required_pages_per_second_upper_estimate": 0.0017046451395841353
+      "required_token_positions_per_second": 27993.508000816706,
+      "required_pages_per_second_upper_estimate": 0.001704657766664014
     },
     {
       "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": 27993.300641739403,
-      "required_pages_per_second_upper_estimate": 0.0021550017738168827
+      "required_token_positions_per_second": 27993.508000816706,
+      "required_pages_per_second_upper_estimate": 0.00215501773689266
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 27993.300641739403,
-      "required_pages_per_second_upper_estimate": 0.347583106419899
+      "required_token_positions_per_second": 27993.508000816706,
+      "required_pages_per_second_upper_estimate": 0.34758568112565286
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05369269309369429,
-    "producer_p95_seconds": 0.18286072099999728,
-    "gpu_update_seconds": 0.9876648828889959,
-    "producer_batches_per_second": 18.624508147784468,
-    "gpu_batches_per_second": 1.012489172516616,
-    "producer_headroom": 18.394772658646694,
+    "producer_batch_seconds": 0.07310230539843587,
+    "producer_p95_seconds": 0.1847567569993771,
+    "gpu_update_seconds": 0.9876575668613371,
+    "producer_batches_per_second": 13.679459143588055,
+    "gpu_batches_per_second": 1.0124966724832432,
+    "producer_headroom": 13.510621333735248,
     "sustainable": true,
-    "buffer_seconds": 60.24755785622875,
+    "buffer_seconds": 60.24711157854156,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374416.8746565459,
-    "upload_bytes_per_second": 5971690.054112959,
+    "generation_bytes_per_second": 374360.9445798159,
+    "upload_bytes_per_second": 5980853.179529553,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.34114532253534,
+    "estimated_full_shards_per_hour": 80.32914403005465,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 9626721,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9389861879999444,
+      "bytes": 6177077,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 88600, "loss": 0.6345845386385918, "lr": 0.0003, "grad_norm": 0.13977763920575317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12966471919, "tokens_per_second": 144134.3117616288, "data_wait_s": 1.2466000043787062e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 88700, "loss": 0.6238090228289366, "lr": 0.0003, "grad_norm": 0.13314721013217592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12980961945, "tokens_per_second": 147591.8424126812, "data_wait_s": 1.3113000022713095e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 88800, "loss": 0.6018594987690449, "lr": 0.0003, "grad_norm": 0.14175604840705872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12995358103, "tokens_per_second": 145818.12052424427, "data_wait_s": 9.759999556990806e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 88900, "loss": 0.6369328573346138, "lr": 0.0003, "grad_norm": 0.13368805654694568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13010034356, "tokens_per_second": 149093.77848599263, "data_wait_s": 9.379000402987003e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:facc7263d1b5cd36c273b55580919fa65ecbefc1bc43f4b923a8069a9a2b22d3
-size 147090
+oid sha256:af22246ce91287cbc048ff0d1ab373fff95c0420d30eb21689461a6bb1cac4f5
+size 149208