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": 75435.63307180545,
-      "required_pages_per_second_upper_estimate": 0.006538569484800019
+      "required_token_positions_per_second": 74308.35172612713,
+      "required_pages_per_second_upper_estimate": 0.006440859594825072
     },
     {
       "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": 9429.454133975682,
-      "required_pages_per_second_upper_estimate": 0.00376155023694578
+      "required_token_positions_per_second": 9288.543965765892,
+      "required_pages_per_second_upper_estimate": 0.003705339064052135
     },
     {
       "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": 18858.908267951363,
-      "required_pages_per_second_upper_estimate": 0.004362676522341011
+      "required_token_positions_per_second": 18577.087931531783,
+      "required_pages_per_second_upper_estimate": 0.004297482347378861
     },
     {
       "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": 28288.362401927046,
-      "required_pages_per_second_upper_estimate": 0.0017226128527101483
+      "required_token_positions_per_second": 27865.631897297677,
+      "required_pages_per_second_upper_estimate": 0.0016968707828737693
     },
     {
       "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": 28288.362401927046,
-      "required_pages_per_second_upper_estimate": 0.002177716444899356
+      "required_token_positions_per_second": 27865.631897297677,
+      "required_pages_per_second_upper_estimate": 0.0021451734804600548
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28288.362401927046,
-      "required_pages_per_second_upper_estimate": 0.3512467859732427
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27865.631897297677,
+      "required_pages_per_second_upper_estimate": 0.34599788789373426
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.05500868710154805,
-    "producer_p95_seconds": 0.18038477799927932,
-    "gpu_update_seconds": 0.9773630444622902,
-    "producer_batches_per_second": 18.178946866227935,
-    "gpu_batches_per_second": 1.0231612558567362,
-    "producer_headroom": 17.767430854294744,
+    "producer_batch_seconds": 0.3714758702499381,
+    "producer_p95_seconds": 0.188758112000869,
+    "gpu_update_seconds": 0.9921899529104602,
+    "producer_batches_per_second": 2.691964889474989,
+    "gpu_batches_per_second": 1.0078715240631393,
+    "producer_headroom": 2.6709405169248015,
     "sustainable": true,
-    "buffer_seconds": 60.59650875666199,
+    "buffer_seconds": 60.52358712753807,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344531.15385919483,
-    "upload_bytes_per_second": 5646488.902948892,
+    "generation_bytes_per_second": 321805.4892227021,
+    "upload_bytes_per_second": 5466001.978673897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.92836534339794,
+    "estimated_full_shards_per_hour": 69.05196673880384,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 5841398,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1767334710002615,
+      "bytes": 1036234,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -31,3 +31,4 @@
 {"step": 283300, "loss": 0.5941618494689465, "lr": 0.0003, "grad_norm": 0.14795023479137123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41467242427, "tokens_per_second": 150159.80292535602, "data_wait_s": 1.1094000001321547e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 283400, "loss": 0.5975897945463657, "lr": 0.0003, "grad_norm": 0.16067724468399516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41481867567, "tokens_per_second": 150081.94934486123, "data_wait_s": 1.3803999536321498e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 283500, "loss": 0.6088459976017475, "lr": 0.0003, "grad_norm": 0.15685733143738648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41496502680, "tokens_per_second": 150456.54305467394, "data_wait_s": 1.3289998605614528e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 283600, "loss": 0.6398585382848978, "lr": 0.0003, "grad_norm": 0.1532649933921476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41511170096, "tokens_per_second": 149175.92814834038, "data_wait_s": 1.2479000361054204e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:571d5e31cbf73b5f41b2780fe87f536bab24ad1ac37e9cce7834c31ae2c1fa0d
-size 70059
+oid sha256:9d724113e00781ef0348decc268c55197a9aae5bfa3335284b48929a7924ee44
+size 72177