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": 75283.50979174026,
-      "required_pages_per_second_upper_estimate": 0.006525383824436896
+      "required_token_positions_per_second": 75963.91278253675,
+      "required_pages_per_second_upper_estimate": 0.006584359431213521
     },
     {
       "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": 9410.438723967532,
-      "required_pages_per_second_upper_estimate": 0.003753964705587814
+      "required_token_positions_per_second": 9495.489097817093,
+      "required_pages_per_second_upper_estimate": 0.003787892571332812
     },
     {
       "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": 18820.877447935065,
-      "required_pages_per_second_upper_estimate": 0.004353878761449812
+      "required_token_positions_per_second": 18990.978195634187,
+      "required_pages_per_second_upper_estimate": 0.004393228575759091
     },
     {
       "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": 28231.3161719026,
-      "required_pages_per_second_upper_estimate": 0.0017191390366000973
+      "required_token_positions_per_second": 28486.46729345128,
+      "required_pages_per_second_upper_estimate": 0.0017346764012279353
     },
     {
       "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": 28231.3161719026,
-      "required_pages_per_second_upper_estimate": 0.0021733248681980026
+      "required_token_positions_per_second": 28486.46729345128,
+      "required_pages_per_second_upper_estimate": 0.00219296710783833
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28231.3161719026,
-      "required_pages_per_second_upper_estimate": 0.3505384627177893
+      "required_token_positions_per_second": 28486.46729345128,
+      "required_pages_per_second_upper_estimate": 0.35370658571155217
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06151869228899898,
-    "producer_p95_seconds": 0.20328271099970152,
-    "gpu_update_seconds": 0.9793379745970487,
-    "producer_batches_per_second": 16.255221995003687,
-    "gpu_batches_per_second": 1.021097951819394,
-    "producer_headroom": 15.91935618521231,
+    "producer_batch_seconds": 0.3198474377186926,
+    "producer_p95_seconds": 0.2684076389996335,
+    "gpu_update_seconds": 0.9705661188235849,
+    "producer_batches_per_second": 3.126490576671447,
+    "gpu_batches_per_second": 1.0303265080096673,
+    "producer_headroom": 3.034465824538518,
     "sustainable": true,
-    "buffer_seconds": 60.718954425017024,
+    "buffer_seconds": 60.175099367062266,
     "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": 550668.4825541043,
-    "upload_bytes_per_second": 4614129.236412751,
+    "generation_bytes_per_second": 491836.5650879355,
+    "upload_bytes_per_second": 4712872.648130926,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.1606374499068,
+    "estimated_full_shards_per_hour": 105.53667749861286,
     "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": 1.322158891000072,
-      "bytes": 2479484,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3229940370001714,
       "bytes": 6465970,
@@ -562,6 +556,12 @@
       "bytes": 1118624,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1147228019999602,
+      "bytes": 10226673,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -140,3 +140,4 @@
 {"step": 251100, "loss": 0.5362557414919138, "lr": 0.0003, "grad_norm": 0.14140154783670644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36754252916, "tokens_per_second": 148980.76062488835, "data_wait_s": 1.633200008654967e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 251200, "loss": 0.5882185809314251, "lr": 0.0003, "grad_norm": 0.17080860579262006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36768800713, "tokens_per_second": 149108.2369015129, "data_wait_s": 1.3812001270707697e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 251300, "loss": 0.6106075681746006, "lr": 0.0003, "grad_norm": 0.15016692406656512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36783446647, "tokens_per_second": 148903.80038179146, "data_wait_s": 1.3144999684300274e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 251400, "loss": 0.5768740810453892, "lr": 0.0003, "grad_norm": 0.13527486253268736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36797985847, "tokens_per_second": 149063.8087277847, "data_wait_s": 1.662700015003793e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:81b50bfb75726a3d7ca77bc8d3584207eeb5acc2a050f5a150a445e0fe4d66d5
-size 303270
+oid sha256:3fa2ab4b10df27d862840295455b8aefd26316ed6b3f9349236dc1ec32395b2e
+size 305388