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": 75585.15262414842,
-      "required_pages_per_second_upper_estimate": 0.006551529460643267
+      "required_token_positions_per_second": 75083.00717074238,
+      "required_pages_per_second_upper_estimate": 0.006508004765418047
     },
     {
       "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": 9448.144078018553,
-      "required_pages_per_second_upper_estimate": 0.003769005935064047
+      "required_token_positions_per_second": 9385.375896342797,
+      "required_pages_per_second_upper_estimate": 0.003743966768925641
     },
     {
       "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": 18896.288156037106,
-      "required_pages_per_second_upper_estimate": 0.004371323701586086
+      "required_token_positions_per_second": 18770.751792685594,
+      "required_pages_per_second_upper_estimate": 0.004342283073289249
     },
     {
       "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": 28344.432234055657,
-      "required_pages_per_second_upper_estimate": 0.0017260272113111064
+      "required_token_positions_per_second": 28156.127689028395,
+      "required_pages_per_second_upper_estimate": 0.0017145604524765406
     },
     {
       "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": 28344.432234055657,
-      "required_pages_per_second_upper_estimate": 0.002182032855787846
+      "required_token_positions_per_second": 28156.127689028395,
+      "required_pages_per_second_upper_estimate": 0.002167536650651305
     },
     {
       "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": 28344.432234055657,
-      "required_pages_per_second_upper_estimate": 0.3519429856346233
+      "required_token_positions_per_second": 28156.127689028395,
+      "required_pages_per_second_upper_estimate": 0.3496048734001564
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.09579807103115456,
-    "producer_p95_seconds": 0.20420349099913437,
-    "gpu_update_seconds": 0.9754296636353541,
-    "producer_batches_per_second": 10.43862354676003,
-    "gpu_batches_per_second": 1.025189244576666,
-    "producer_headroom": 10.182143055032224,
+    "producer_batch_seconds": 0.508773420999816,
+    "producer_p95_seconds": 0.22332594900035474,
+    "gpu_update_seconds": 0.9819532112284604,
+    "producer_batches_per_second": 1.965511480601424,
+    "gpu_batches_per_second": 1.018378460974696,
+    "producer_headroom": 1.930040310082974,
     "sustainable": true,
-    "buffer_seconds": 60.47663914539196,
+    "buffer_seconds": 60.88109909616454,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316080.96527384076,
-    "upload_bytes_per_second": 5826394.503146824,
+    "generation_bytes_per_second": 280631.6634791335,
+    "upload_bytes_per_second": 5764616.0707132965,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.82361715947549,
+    "estimated_full_shards_per_hour": 60.217022211842576,
     "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": 1.1579725150004379,
-      "bytes": 10569380,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.112220934999641,
       "bytes": 6262678,
@@ -562,6 +556,12 @@
       "bytes": 7770049,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1648699769993982,
+      "bytes": 6106287,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -117,3 +117,4 @@
 {"step": 366700, "loss": 0.6004631537944078, "lr": 0.0003, "grad_norm": 0.16477317280781276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53674931778, "tokens_per_second": 149537.63814292118, "data_wait_s": 1.3441000191960484e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 366800, "loss": 0.5651937630027533, "lr": 0.0003, "grad_norm": 0.1773074711262574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53689494372, "tokens_per_second": 149013.9721889122, "data_wait_s": 1.0245999874314293e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 366900, "loss": 0.6010194253176451, "lr": 0.0003, "grad_norm": 0.144458527558294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53704024381, "tokens_per_second": 145445.4375571416, "data_wait_s": 1.1977999747614376e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 367000, "loss": 0.5593036767095327, "lr": 0.0003, "grad_norm": 0.17470027523121276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53718637757, "tokens_per_second": 149394.50868001793, "data_wait_s": 1.003299985313788e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7107886ed039db369ae084ac9892c44d0df94417ba80995a86a5c807f78895e1
-size 253773
+oid sha256:3400140e6995c7a04c1915566d2a516a79ad48b68e03e723d50dacbe4a6c551f
+size 255891