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": 75836.36166497172,
-      "required_pages_per_second_upper_estimate": 0.006573303623618337
+      "required_token_positions_per_second": 75741.78948840455,
+      "required_pages_per_second_upper_estimate": 0.006565106347044485
     },
     {
       "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": 9479.545208121464,
-      "required_pages_per_second_upper_estimate": 0.0037815323153508316
+      "required_token_positions_per_second": 9467.723686050569,
+      "required_pages_per_second_upper_estimate": 0.003776816533449246
     },
     {
       "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": 18959.09041624293,
-      "required_pages_per_second_upper_estimate": 0.004385851899202673
+      "required_token_positions_per_second": 18935.447372101138,
+      "required_pages_per_second_upper_estimate": 0.004380382497043835
     },
     {
       "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": 28438.635624364393,
-      "required_pages_per_second_upper_estimate": 0.001731763703533917
+      "required_token_positions_per_second": 28403.171058151707,
+      "required_pages_per_second_upper_estimate": 0.0017296040975197631
     },
     {
       "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": 28438.635624364393,
-      "required_pages_per_second_upper_estimate": 0.002189284893545487
+      "required_token_positions_per_second": 28403.171058151707,
+      "required_pages_per_second_upper_estimate": 0.002186554733066232
     },
     {
       "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": 28438.635624364393,
-      "required_pages_per_second_upper_estimate": 0.3531126764637917
+      "required_token_positions_per_second": 28403.171058151707,
+      "required_pages_per_second_upper_estimate": 0.352672325243698
     }
   ],
   "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": 0.10543953207809409,
-    "producer_p95_seconds": 0.3091674610004702,
-    "gpu_update_seconds": 0.972198538818542,
-    "producer_batches_per_second": 9.484108856432966,
-    "gpu_batches_per_second": 1.0285964852562353,
-    "producer_headroom": 9.220436772220124,
+    "producer_batch_seconds": 0.14954435957815804,
+    "producer_p95_seconds": 0.3460210890007147,
+    "gpu_update_seconds": 0.9734124384701414,
+    "producer_batches_per_second": 6.686979053043848,
+    "gpu_batches_per_second": 1.0273137680176399,
+    "producer_headroom": 6.509188586022169,
     "sustainable": true,
-    "buffer_seconds": 60.276309406749604,
+    "buffer_seconds": 60.351571185148764,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329443.6854598167,
-    "upload_bytes_per_second": 4429483.606356619,
+    "generation_bytes_per_second": 315490.2883598089,
+    "upload_bytes_per_second": 4376808.522885742,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.69094584318043,
+    "estimated_full_shards_per_hour": 67.69687164397908,
     "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": 1.2721305819977715,
-      "bytes": 12594000,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.3089486960016075,
       "bytes": 6052165,
@@ -562,6 +556,12 @@
       "bytes": 930296,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.2613436630017532,
+      "bytes": 7588531,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -280,3 +280,4 @@
 {"step": 265100, "loss": 0.5947478972375393, "lr": 0.0003, "grad_norm": 0.16243270863430434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38803747659, "tokens_per_second": 124333.38661100768, "data_wait_s": 1.1961998097831383e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 265200, "loss": 0.5762580297887325, "lr": 0.0003, "grad_norm": 0.15203765024384844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38818274096, "tokens_per_second": 148059.59847431353, "data_wait_s": 1.0679999832063913e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 265300, "loss": 0.6618193089962006, "lr": 0.0003, "grad_norm": 0.1676675397491687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38832863645, "tokens_per_second": 148181.58215147804, "data_wait_s": 1.0504998499527574e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 265400, "loss": 0.5782671216875315, "lr": 0.0003, "grad_norm": 0.15107701337907797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38847406928, "tokens_per_second": 148437.16574777808, "data_wait_s": 1.2975000572623685e-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 sha256:8c837ffc9f4aa5302592433c3fb4d2feccdd02412d2577945894188fb29a86e8
-size 602139
+oid sha256:de709438c7bdcc54ca281b1b37d56c70de64a91ac5f48b4b8c2966d4d8b3c6a2
+size 604257